The Role of Visual Cues in Online Reviews: How Image Complexity Shapes Review Helpfulness

Abstract

Online reviews play a critical role in shaping consumer decisions and providing valuable insights to enhance the products and services for businesses. As visual content becomes increasingly prevalent in reviews, it is essential to understand how image complexity influences review helpfulness. Despite the growing importance of images, the impact of color diversity and texture homogeneity on review helpfulness remains underexplored. Grounded in Information Diagnosticity Theory and Dual Coding Theory, this study investigates the relationship between image complexity and review helpfulness, as well as the moderating role of review text readability. Using a large-scale dataset from the hotel and travel sectors, the findings reveal that color diversity has a positive effect on review helpfulness, while texture homogeneity follows an inverted U-shaped relationship with helpfulness. Furthermore, text readability strengthens the positive impact of texture homogeneity, making moderately homogeneous images more effective when paired with clear and well-structured text. Heterogeneity analysis demonstrates that these effects vary across product categories. The results advance the understanding of multimodal information processing in online reviews, providing actionable guidance for platforms and businesses to refine the review systems.

1. Introduction

In the digital era, consumers increasingly rely on online information to make purchasing decisions. Imagine a traveler planning a vacation and choosing between two comparable hotels on an online platform. While one hotel provides numerous text-based reviews, the other offers multimodal reviews enriched with vibrant images that showcase room interiors, amenities, and surrounding landscapes. The visual cues displayed by these images can significantly influence travelers’ perceptions, potentially tipping the decision in favor of the hotel with image-enhanced reviews. This scenario highlights the significant influence of visual information on shaping consumer perceptions and decisions within online environments.

Visual information has a profound impact on consumer behavior, as it enhances understanding, reduces uncertainty, and influences emotions. Studies have shown that images in online content can increase engagement, improve memory recall, and affect purchase intentions [1]. In online reviews, visual cues, such as images, complement textual information by providing richer representations of products or services, thereby making the reviews more helpful [2,3]. From the perspective of photography, these images in reviews provide specific information about the target products or services, including associated components within the same context [4]. The complexity of these images, encompassing elements, such as color diversity and size, can either enhance or impede their effectiveness [5]. While some research suggests that complex images can capture attention and convey detailed information [6], others argue that excessive complexity may lead to cognitive overload, reducing information processing efficiency [7].

Online reviews are now indispensable in the decision-making processes, acting as a form of electronic word-of-mouth (eWOM) that significantly influences perceptions of product quality, credibility, and trustworthiness [8]. With the exponential growth of e-commerce, consumers increasingly rely on the experiences and opinions of others to guide their purchasing decisions, particularly in environments characterized by information asymmetry and perceived risk [9,10]. Helpful online reviews not only affect individual consumer choices but also have a measurable impact on product sales and brand reputation [11].

The helpfulness of reviews reflects the extent to which they assist consumers in making informed decisions by providing valuable information [12]. Consumers often trust reviews that have been deemed helpful by others, using these endorsements as a heuristic for credibility and utility [2]. Extensive research has explored factors affecting review helpfulness, focusing primarily on textual characteristics such as review length, sentiment, extremity, and linguistic style [2,12,13,14]. For instance, longer reviews tend to be perceived as more helpful due to the detailed information they provide [15]. Reviews with moderate ratings are often seen as more credible, as they appear more balanced and less biased [12]. However, despite the growing prevalence of images in online reviews, limited attention has been given to how image complexity influences perceived helpfulness. The research gaps give rise to the following unanswered research questions. What visual cues influence review helpfulness? How do visual cues influence review helpfulness? Is this influence affected by the characteristics of the text review? As platforms increasingly incorporate visual content, understanding its impact on review helpfulness becomes essential for both theory and practice.

To address this gap, the present study examines how visual cues (color diversity and texture homogeneity) and textual cues (readability) interact to influence consumers’ evaluations of online reviews. Drawing on Information Diagnosticity Theory and Dual Coding Theory, we develop a research framework and then analyze online review data in the hotel and travel sectors, to examine how these visual attributes influence review helpfulness. This research contributes to the literature by integrating visual factors into the analysis of review helpfulness, offering practical insights for consumers, platform managers, and marketers to enhance the effectiveness of online reviews.

The rest of the paper is organized as follows. Section 2 reviews the related research work. Section 3 points out the fundamental theories used to elaborate the hypotheses. Section 4 describes the empirical data and variable descriptions. The empirical analysis, including the empirical model, result analysis, robustness test, and heterogeneity analysis, is demonstrated in Section 5. Finally, Section 6 presents the discussion, including theoretical and managerial implications, limitations, and further research.

2. Related Works

2.1. Review Helpfulness in Consumer Decision-Making

2.1.1. Importance of Online Reviews

Before making an online purchase, consumers tend to seek extensive information to evaluate the products or services thoroughly, then determine their suitability for personal needs. However, information presented by sellers tends to be positively biased and broadly promotional, which may lead consumers to view it with skepticism. As a result, potential buyers are increasingly relying on reviews that reflect authentic consumer experiences. With the continued development of e-commerce platforms, user-generated content has become a crucial information source for online shoppers, particularly in the form of online reviews [8]. Studies indicate that over 92% of consumers consult relevant reviews before making a final purchase decision, underscoring the substantial influence of online reviews [2]. Potential customers are more receptive to reviews they perceive as helpful. Consequently, the more helpful the review, the stronger its influence on a consumer’s purchasing decision [10].

The helpfulness of reviews reflects their diagnostic value, meaning consumers can accurately assess the quality of products or services based on the information provided [16]. Helpful reviews have a significant influence on consumers’ decision-making [10]. Most studies measure the helpfulness of reviews by utilizing the platform’s “usefulness” voting system, and the most common metrics of usefulness are the number of helpful votes [17] and the proportion of helpful votes [18]. The subjective nature of reviews can lead to inconsistencies and biases in the voting system, leaving potential consumers uncertain about why specific ratings are assigned a high or low value [2]. Additionally, reviews posted later may receive fewer votes, leading to an incorrect assessment of their value [19]. While review helpfulness is challenging to measure accurately, its influence on consumer behavior remains significant.

2.1.2. Factors Affecting Review Helpfulness

Scholars have examined a range of factors that influence review helpfulness, these factors can be broadly classified into two main categories: reviewer-related factors and review-related factors.

Reviewer-related factors mainly consist of personal information disclosure, professionalism, and credit rating. Disclosing personal details, such as registration date, name, and location can enhance the credibility of reviews [15]. The professionalism of reviewers also serves as a key determinant of review helpfulness, as professional reviewers typically produce higher-quality content [20]. Similarly, reviewers with high credit ratings are often seen as more trustworthy, where the reputational credibility can be described as the social status or expert image of the information source [21]. As a reviewer gains more attention, their reputation increases. This rise in reputation not only enhances the visibility of a review’s content but also bolsters its perceived credibility and influence, making it more valuable to consumers [22].

Review-related factors primarily contain the content of the text, the time of publication, and the rating. Text content can be divided into intrinsic and extrinsic features. Intrinsic features include elements such as emotion, language style, and the proportion of parts of speech used in the review. The emotion conveyed in reviews or inferred from the emotional tone of the text, reflects consumers’ attitudes towards the product and can resonate with other consumers, thereby increasing the perceived value of the review [13,20]. Empirical evidence has shown that extreme reviews are more influential than moderate reviews because the former provide more explicit information while the latter often provide more vague information [2]. Additionally, potential buyers generally prefer reviews that are clearly written and easy to understand [14]. Extrinsic features, such as review length and character count, also significantly impact a review’s helpfulness. Longer reviews often contain more detailed information [15] and are generally perceived as more helpful [23]. Online reviews have evolved beyond simple text and are now presented in multimodal forms combining text, images, and videos. Recent studies have been dedicated to revealing multimodal factors affecting review helpfulness by employing machine learning techniques [24] or AI-based methods, such as multimodal fusion [25] and the BLIP2 model [26]. The different review formats have various effects on the perceived helpfulness of reviews. Among the multimodal factors, text elements contribute the most [23], while both text quality and photo aesthetic quality have a positive influence on review helpfulness [24].

The publication time and rating also affect review helpfulness. Publication time refers to the interval between when a review was posted and when the data were collected, usually measured in days or months [19]. Different scholars hold varying views on the impact of this variable. Luo et al. [20] suggested that reviews published earlier are more likely to be read and deemed helpful by consumers, while Yin et al. [27] argued that some websites display comments in chronological order, which may lead to earlier reviews being overshadowed by more recent ones, potentially diminishing their perceived helpfulness. A rating represents a consumer’s overall evaluation of a product, often expressed on a scale (e.g., 1–5 stars) or as a simple “good” or “bad” assessment. Some scholars propose a “negative bias”, where lower-rated reviews are more likely to attract consumer attention and are thus perceived as more helpful [28]. Other studies suggest that reviews with moderate ratings tend to be more helpful than those with extreme positive or negative ratings [12]. Apart from the above factors, the type of product or service also affects review helpfulness as consumers want to obtain various information according to product features [29].

2.2. The Influence of Visual Cues on Consumer Behavior

Research in advertising and psychology suggests that the visual or pictorial component of advertisements is encoded more elaborately and distinctly than the verbal component, thereby helping consumers better understand the presented information [30]. Image reviews can convey more specific product features, thereby enriching the overall content of reviews. Current research on image reviews primarily focuses on factors such as the presence or absence of images [3], the image count [13], and their characteristics. A higher number of images can more effectively support reviewers’ opinions, making the reviews more credible [4,13].To gain a deeper understanding of the role of images, deep neural network-based methods are used to explore the value of visual elements [31]. Lei et al. [32] integrated convolutional neural networks and long short-term memory algorithms to interpret image content, while Sun and Liu [33] developed a visual-semantic embedding model to investigate the impact of user-generated images on product sales. These AI-driven image understanding methods provide a technical foundation for the automated evaluation of image effectiveness.

Compared to plain text reviews, image reviews can compensate for insufficient product information, enhancing consumers’ perceived helpfulness [2,13]. Krause [34] categorized image features into explicit and implicit characteristics, in which explicit features are objective and directly observable without ambiguity, such as color, shape, size, and clarity. Chi et al. [35] confirmed that the hue and brightness in image reviews affect the rental purchase of hotels. Implicit features, on the other hand, require subjective interpretation, including emotional tones or symbolic meanings [4]. These image characteristics often reflect the diverse information contained within a picture, helping consumers assess product information from multiple perspectives [18,36], which in turn influences their purchasing decisions. In other words, images not only reflect the quality and reliability of review information, but they also enhance consumers’ purchase intentions [2].

“A picture is worth a thousand words.” The images can be considered a more informative data source than text alone [8]. Despite the growing prevalence of image reviews in practice, academic research in this area remains relatively limited. Most existing studies focus on structural aspects, such as the presence and quantity of images, while research on the content and visual features of images themselves remains scarce. This gap suggests that our understanding of how visual elements in image reviews affect review helpfulness is still insufficient. This study addresses the gap by examining how image complexity affects review helpfulness, thereby offering new insights into the field.

3. Research Design and Hypotheses Development

3.1. Theoretical Framework

In this study, we employ two complementary theories—Information Diagnosticity Theory and Dual Coding Theory—to explain the expected relationships between visual cues and review helpfulness, as well as the moderating role of textual readability.

Information Diagnosticity Theory posits that consumers evaluate information sources based on their ability to reduce uncertainty and facilitate informed decision-making [37]. Diagnostic information enables consumers to make more accurate product judgments by offering clarity and reliability. Prior research has primarily focused on verbal content, such as sentiment strength and argument quality, as key determinants of diagnosticity [38]. This perspective has been expanded by recognizing the role of images in enhancing perceived informativeness [39]. The recent literature suggests that perceived diagnosticity plays a crucial role in shaping review helpfulness. Piva and Marques [40] argue that online book reviews are perceived as more useful when they effectively reduce consumer uncertainty. Additionally, Islam et al. [41] highlight that review credibility increases when it contains diagnostic cues such as detailed descriptions and authentic user-generated content. These findings reinforce the importance of integrating both textual and visual elements to enhance the informativeness of online reviews.

Dual Coding Theory asserts that individuals process information through two cognitive channels: a verbal channel for textual content and a non-verbal channel for visual stimuli [42]. When both channels are activated simultaneously, cognitive processing and retention improve due to the dual encoding of information. In the context of online reviews, textual comments engage the verbal channel, while images trigger the non-verbal channel, leading to better comprehension and recall [43]. Recent research underscores the role of multimodal presentation in consumer decision-making. Vazquez et al. [44] find that reviews featuring both high-quality images and clear textual explanations significantly enhance consumer trust and engagement. However, when textual readability is low, cognitive load increases, reducing consumers’ ability to process accompanying visual information effectively [45]. Additionally, product presentation videos increase perceived diagnosticity by engaging both verbal and visual processing pathways [39].

Information Diagnosticity Theory and Dual Coding Theory provide complementary perspectives on how consumers process online reviews. Information Diagnosticity Theory emphasizes that consumers evaluate information based on its ability to reduce uncertainty and facilitate decision-making. In online reviews, clear and detailed visual cues enhance informativeness, influencing perceived reliability. Dual Coding Theory explains how consumers process information through verbal and non-verbal channels, suggesting that integrating text and images improves comprehension and retention. Together, these theories provide a robust framework for understanding how visual and textual elements interact to shape the helpfulness of reviews and consumer perceptions.

3.2. The Effect of Image Complexity on Review Helpfulness

3.2.1. The Effect of Color Diversity on Review Helpfulness

Color diversity refers to the range of distinct colors present in a review image. When an image has greater color diversity, it tends to convey more nuanced visual cues [46]. Thus, higher color diversity can increase the perceived richness of the review and help readers learn more about the product, thereby boosting the review’s helpfulness [47].

From the perspective of Information Diagnosticity Theory, color diversity increases perceived informativeness because consumers believe they have more cues to reduce uncertainty about product quality or appearance [12]. Moreover, according to Dual Coding Theory [42], higher color diversity can stimulate the visual processing channel more effectively, leading to better comprehension and recall of the product details shown in the image. Even if the text section of a review is not exceptionally detailed, a colorful image can enhance the overall perceived helpfulness of the review by offering complementary visual evidence [47].

In addition, color diversity may also create a sense of vividness. The “vividness effect” [48] implies that more colorful, eye-catching presentations stand out from other reviews, potentially drawing more attention and encouraging readers to vote “helpful” [49]. Hence, we propose the following:

H1: 

Color diversity in review images is positively associated with review helpfulness.

3.2.2. The Effect of Texture Homogeneity on Review Helpfulness

Texture homogeneity is another important aspect of image complexity. Texture relates to the overall arrangement of pixels, patterns, or surfaces in an image [50]. In online reviews, a moderately homogeneous texture may convey an image that is clear and not overly noisy, making it easy for viewers to identify the product’s shape, material quality, or other fine details [51,52]. When texture has a medium level of uniformity, a good balance is achieved between clarity and detail, thereby enhancing diagnostic value [12].

However, if texture homogeneity is too low, the image may look chaotic or blurry, preventing consumers from grasping what the product truly looks like [52]. Conversely, if texture homogeneity is too high, the image might appear overly uniform, smoothing out relevant details or making the product look artificial. When consumers cannot see subtle imperfections or genuine textures, they may doubt the image’s authenticity, thus lowering its credibility [53].

For the online review images, a moderately homogeneous texture suggests a clear but detailed view of the product [54]. If the texture is highly uniform, it may lose diagnostic value, as there are fewer distinctive features to confirm the product’s information. This leads to the expectation of an inverted U-shaped relationship between texture homogeneity and review helpfulness, wherein moderate homogeneity yields the maximum helpfulness. Thus, the following hypothesis is constructed:

H2: 

Texture homogeneity in review images demonstrates an inverted U-shaped relationship with review helpfulness. Specifically, review helpfulness increases as texture homogeneity moves from low to moderate levels and then decreases as homogeneity becomes very high.

3.3. The Moderating Role of Review Text Readability

While the above sections focus on image complexity (color diversity and texture homogeneity), we also consider text readability as a potential moderator. Readability refers to the ease with which a reader can process and comprehend the textual content of a review [55,56]. Previous studies have shown that the readability of the review text enhances the helpfulness of the review [57,58,59].

According to Dual Coding Theory [42], verbal and visual channels can reinforce each other. If the textual portion of a review is highly readable, readers are more inclined to absorb the text quickly and then engage with the accompanying image, or vice versa. When the text is difficult to read, readers may expend extra cognitive effort deciphering it, which leaves fewer cognitive resources for analyzing the visual cues [60]. As a result, the benefit of color diversity becomes less pronounced in a review written with complex and confusing language. Moreover, readers may form a general impression of quality based on the textual clarity, which can influence their perception of the image [61]. A clear, concise textual explanation can make the varied colors in the image more meaningful and easier to interpret. This suggests that readability should strengthen the positive effect of color diversity on review helpfulness. Accordingly, we hypothesize the following.

H3: 

Text readability positively moderates the relationship between color diversity and review helpfulness. In other words, the positive influence of color diversity on review helpfulness is more substantial when text readability is high.

Similarly, we propose that text readability influences how consumers interpret or benefit from different levels of texture homogeneity. When the text is more readable, readers can gather essential product information from the text, allowing them to better understand or confirm what they see in the image [60,61]. For example, if the text describes the product as having a smooth surface, then a moderately homogeneous texture in the image will serve as visual evidence of that description. In contrast, when readability is low, readers may not fully grasp the textual details, making it harder for them to evaluate images at intermediate or high levels of homogeneity. They might misinterpret a highly uniform image or fail to recognize its connection to the textual explanation. As a result, the optimal (moderate) point of homogeneity may become less clear to the consumer if the text is not understandable [62]. With a high level of textual clarity, however, consumers can interpret moderate homogeneity more effectively, perceiving it as credible and useful [61].

Hence, we predict that high readability will strengthen the benefits of moderate texture homogeneity and possibly reduce the negative perception when homogeneity is at extreme levels, because the text can clarify or supplement missing details. The following hypothesis is thus developed.

H4: 

Text readability positively moderates the inverted U-shaped relationship between texture homogeneity and review helpfulness. Specifically, the peak effect of texture homogeneity is more substantial and potentially broader under high readability conditions than under low readability conditions.

4. Empirical Data and Variables Description

4.1. Data Description

The datasets used to test the proposed hypotheses focus on hotels and travel, selected for two primary reasons. First, with the rapid growth of the experience economy, consumer demand for services has steadily increased, positioning travel as a key avenue for leisure and entertainment. In planning their trips, consumers often prioritize the comfort of their accommodations. Unlike traditional hotels, homestays offer personalized and unique experiences, making them a preferred choice for travelers seeking distinctive lodging options. Second, both hotels and travel generate substantial volumes of online reviews, reflecting a level of consumer interest. Research has shown that images of destinations and restaurants in online reviews significantly influence consumers’ intentions to visit or make a purchase in the tourism and hospitality industry [63,64]. This abundance of user-generated content provides a rich dataset for empirical analysis.

Data were sourced from two major online travel platforms in China, Ctrip and Tongcheng, both known for their extensive user bases, transaction data, and structured review systems, ensuring the availability of high-quality data. For the hotel segment, reviews were collected from Ctrip for accommodations in Shanghai, Hangzhou, and Ningbo. A total of 42,333 online reviews were gathered from these cities. For the travel segment, 43,615 reviews were collected from Tongcheng.

The data collection process captured various variables, including review content, number of reviews, ratings, images, publication dates, and reviewer names. After data cleaning, 14,111 image-containing reviews were extracted from the 42,333 hotel reviews, and 11,010 image-containing reviews were selected from the 43,615 travel reviews. The final dataset comprised 25,121 image-based reviews, serving as the basis for further analysis.

4.2. Variable Design

4.2.1. Dependent Variable

The dependent variable in this study is review helpfulness. On both Ctrip and Tongcheng platforms, users indicate a review’s helpfulness by clicking a “like” button. This mechanism enables consumers to signal the helpfulness of a review, thereby helping others to filter for more informative content. Accordingly, the number of votes a review receives, i.e., Vote Count, serves as the measure of its helpfulness.

4.2.2. Independent Variables

The independent variables in this study are image complexity, which is composed of two components: color diversity and texture homogeneity. These two variables are derived from image data using computational techniques, providing critical insight into how visual elements in reviews influence their perceived helpfulness.

Color diversity refers to the variety of distinct colors present in an image. In this study, we use color histogram entropy, which quantifies the distribution of colors. First, convert the image to the HSV or Lab color space to focus on perceptual differences. Compute the color histogram using the Hue (H) channel with N bins. Normalize the histogram to get the probability $p_i$ of each color bin. Then, calculate entropy using Shannon’s formula:

$$Color Diversity=-{\displaystyle \sum _{i=1}^N{p}_i{\log }_2(p_i)}$$

(1)

The value of color diversity is calculated using Python, which effectively quantifies the richness of colors in images. A higher value of color diversity indicates a greater variety of distinct colors within an image. Images with high color diversity typically include vibrant and complex scenes, such as a scenic landscape that include the sky, trees, water, and people, or a food photograph of assorted hotpot ingredients. In contrast, images with low color diversity are often monochromatic or grayscale, such as minimalist product shots with a uniform color background. Figure 1 illustrates the sample images, where (a) and (c) show the images of high color diversity, while the remaining two are the ones of low color diversity.

Texture homogeneity measures the intricacy of texture patterns within an image, reflecting the visual surface characteristics such as roughness or smoothness. This study calculates texture homogeneity based on the gray-level co-occurrence matrix (GLCM), which analyzes the frequency with which pairs of pixel values occur in a specific spatial relationship. This value is obtained by Python. The formula for texture homogeneity is as follows:

$$Texture Complexity={\displaystyle \sum _{i,j}\frac{P\left(i,j\right)}{1+\left|i-j\right|}}$$

(2)

where $P(i,j)$ is the normalized GLCM value at the pixel pair $(i,j)$, and $(i-j)$ represents the difference between the pixel values. Higher texture homogeneity reflects visual smoothness and simplicity in the image, characterized by fewer intricate patterns and reduced surface variation. Examples of high texture homogeneity images include a white hotel bedsheet or a clean, uncluttered floor. Conversely, low texture homogeneity is found in images with rich surface details and complex textures, such as patterned fabrics, brick walls, or busy scenes with overlapping visual elements. Figure 1a,b illustrates the sample images of low texture homogeneity, while the rest are the ones of high texture homogeneity. For each review, the values for color diversity and texture homogeneity are averaged across all included images to represent the overall complexity of the image.

It is essential to note that color diversity and texture homogeneity provide objective and replicable measures of image complexity; however, they may not always align with the perceived visual quality by consumers. For example, a grainy or overexposed photo may score high in complexity but offer little diagnostic value. Therefore, we caution that these metrics should be interpreted as proxies rather than direct reflections of perceived informativeness or clarity.

4.2.3. Moderator Variable

Readability of Chinese text is measured using the ‘cntext’ package (https://github.com/hidadeng/cntext (accessed on 4 March 2024), which adapts principles from the Fog Index [65] for Chinese texts. The algorithm incorporates two key components. The first component, denoted as $a$, is the average number of characters per sub-sentence. This metric captures sentence length and reflects the density of information within each segment of the text. The second component, denoted as $b$, is the proportion of adverbs and conjunctions present in each sentence, serving as an indicator of syntactic complexity. These two components are integrated into a single readability score using the following formula:

$$Readability=(a+b)\times 0.5$$

(3)

In this formulation, a higher readability score indicates that the text is more complex and imposes a greater cognitive load on the reader, suggesting that the material is more difficult to understand. Conversely, a lower score implies that the text is simpler and easier to comprehend. This objective measure of textual clarity is crucial for our study, as it serves as a moderator variable to investigate whether the ease of processing review texts influences how consumers integrate visual cues with textual information in evaluating review helpfulness.

4.2.4. Control Variables

Consumers consider multiple factors when evaluating the helpfulness of reviews. Prior research has demonstrated that variables such as rating score, sentiment score, sentence count, image count, word count, review lifespan, and product type, significantly influence review helpfulness.

The rating score refers to the numerical score assigned to a product or service by reviewers (typically on a scale of 1 to 5), which shapes consumers’ perceptions. Moderate or negative ratings are frequently viewed as more objective, while highly positive reviews may be perceived as subjective [12,28].

The sentiment score measures the overall emotional tone of the review text. It quantifies the emotional tone of a review based on a Chinese sentiment lexicon while accounting for the influence of degree adverbs. Reviews with negative sentiments are often regarded as more credible and influential in consumer decision-making compared to positive reviews [17].

Sentence count refers to the total number of sentences in a review text. Higher sentence counts often indicate more comprehensive content, which may influence consumers’ perceptions of helpfulness [20]. Thus, sentence count is included as a control variable to account for length.

Image count refers to the total number of images included in a review. Reviews with more images generally provide richer information, making them more helpful [13].

Word count refers to the total number of words in a review, while sentence count reflects the number of sentences in a review. Longer reviews, with more detailed content, are generally perceived as more helpful [15].

Review lifespan refers to the period between the publication of the review and its inclusion in a collection. Older reviews are often considered more credible due to their longevity and visibility [20].

Product type refers to a categorical variable that indicates the category of the product or service being reviewed, such as hotels or travel. This variable is critical for controlling potential differences in review characteristics and consumer evaluations across distinct product categories [29].

To ensure the robustness of the results, these control variables are incorporated into the regression models to account for their potential influence on review helpfulness, as supported by the prior literature. All the variables used in the study are listed in

Table 1. Description of variables.

Variable Name	Variable Type	Variable Definition
Dependent Variable
Review Helpfulness	Continuous	The number of “likes” voted by consumers, i.e., Vote Count.
(Vote Count)
Independent Variables
Color Diversity	Continuous	The average number of distinct colors across all images in a review.
Texture Homogeneity	Continuous	The average intricacy of texture patterns across all images in a review.
Moderator Variable
Readability	Continuous	A measure of the complexity of Chinese review text, calculated using the ‘cntext’ package
Control Variables
Rating Score	Continuous	The numerical score (typically ranging from 1 to 5) assigned by reviewers to a product or service.
Sentiment Score	Continuous	The overall emotional tone of the review text.
Sentence Count	Continuous	The total number of sentences in a review text.
Image Count	Continuous	The total number of images included in a review.
Word Count	Continuous	The total number of words in a review.
Review Lifespan	Continuous	The time difference in days between the publication of the review and its collection.
Product Type	Dummy	A binary variable indicating the category of the product or service being reviewed (0 for hotels, 1 for travel).

.

4.3. Descriptive Analysis

To understand the basic information about the variables, we present the descriptive statistics for these variables in

Table 2. Descriptive statistics of variables.

Variables	Mean	Std	Min	25%	50%	75%	Max
Vote Count	1.02	5.82	0.00	0.00	0.00	1.00	299.00
Color Diversity	234.84	32.38	235.00	235.00	243.75	249.75	256.00
Homogeneity	0.34	0.14	0.05	0.24	0.32	0.39	0.99
Readability	27.13	29.89	11.50	11.50	18.50	31.00	532.50
Rating Score	4.70	0.42	1.00	4.50	4.70	5.00	5.00
Sentiment Score	4.24	5.05	−21.00	1.00	3.00	6.00	103.00
Sentence Count	3.78	4.24	1.00	1.00	2.00	5.00	54.00
Image Count	4.95	3.00	2.00	2.00	3.00	8.00	15.00
Word Count	97.33	101.54	33.00	33.00	61.00	134.00	809.00
Review Lifespan	603.47	276.60	0.00	397.00	602.00	826.00	1201.00
Product Type	0.44	0.50	0.00	0.00	0.00	1.00	1.00

, including the mean, standard deviation, minimum, and maximum values.

Before conducting the empirical analysis, the data were preprocessed to ensure consistency and reliability. We examined the distribution of each variable. The analysis revealed that the moderator variable, Readability, exhibited significant right-skewed distributions. To mitigate the influence of extreme values and outliers, logarithmic transformations were applied to Readability. Additionally, to eliminate dimensional differences among variables and enhance model stability, Z-score standardization was applied to all variables except the dependent variable. This standardization enables direct comparisons among variables measured on different scales, thereby improving the accuracy and interpretability of the model.

We then examined the correlations among all variables, as shown in

Table 3. Correlation coefficients among variables.

Variables	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)	(10)	(11)
(1) Vote Count	1
(2) Color Diversity	−0.016 ***	1
(3) Homogeneity	0.042 ***	−0.378 ***	1
(4) Readability	0.059 ***	−0.058 ***	0.077 ***	1
(5) Rating Score	−0.084 ***	0.167 ***	−0.276 ***	−0.118 ***	1
(6) Sentiment Score	0.185 ***	−0.102 ***	0.192 ***	0.298 ***	−0.121 ***	1
(7) Sentence Count	0.192 ***	−0.056 ***	0.123 ***	−0.219 ***	−0.201 ***	0.344 ***	1
(8) Image Count	0.117 ***	−0.197 ***	0.186 ***	0.199 ***	−0.170 ***	0.317 ***	0.301 ***	1
(9) Word Count	0.260 ***	−0.087 ***	0.173 ***	0.332 ***	−0.284 ***	0.521 ***	0.712 ***	0.410 ***	1
(10) Review Lifespan	0.063 ***	0.141 ***	−0.068 ***	−0.003	−0.027 ***	0.011 *	0.042 ***	0.007	0.044 ***	1
(11) Product Type	0.121 ***	−0.273 ***	0.472 ***	0.175 ***	−0.502 ***	0.454 ***	0.295 ***	0.395 ***	0.414 ***	0.090 ***	1

Note: * p < 0.1. *** p < 0.01.

, demonstrating that the independent variables are associated with the dependent variable. To assess potential multicollinearity, we conducted a Variance Inflation Factor (VIF) analysis (

Table 4. Multicollinearity test.

Variables	VIF
Color Diversity	2.320631
Homogeneity	2.690006
Readability	2.220405
Rating Score	1.417169
Sentiment Score	1.628837
Sentence Count	3.902577
Image Count	1.351548
Word Count	4.42122
Review Lifespan	1.042133
Product Type	2.162098
Mean VIF	2.31

). The results show an average VIF of 2.31, with no variable exceeding 10, indicating that there is no significant multicollinearity. Thus, the independent variables are suitable for regression analysis.

5. Empirical Analysis

5.1. Empirical Model

To examine the relationship between image complexity and review helpfulness, we construct the following regression model:

$$\begin{array}{l}Review Helpfulness\\ ={\beta }_0+{\beta }_1(Color Diversity)+{\beta }_2(Texture Homogeneity)+{\beta }_3(Readability)\\ +{\beta }_4(Texture Homogeneit{y}^2)+{\beta }_5(Color Diversity\times Readability)\\ +{\beta }_6(Texture Homogeneity\times Readability)\\ +{\beta }_7(Texture Homogeneit{y}^2\times Readability)+Controls+\epsilon \end{array}$$

(4)

where $ReviewHelpfulness$ is measured by the number of helpful votes that a review receives, that is Vote Count. The independent variables include Color Diversity and Texture Homogeneity, representing two key dimensions of image complexity. To account for potential non-linear effects, we include the squared term of Texture Homogeneity. Readability is incorporated as a moderator, and its interaction terms with Color Diversity and Texture Homogeneity (both linear and squared) are included to examine whether textual clarity influences how consumers process visual information.

Given that the dependent variable, Vote Count, exhibits overdispersion (i.e., variance significantly exceeding the mean), a negative binomial regression model is employed as the primary estimation method [66]. This model is well-suited for count data with overdispersion, providing a more robust estimation compared to Poisson regression [67]. We also control for several factors, as illustrated in Table 1, to account for their potential influence on review helpfulness.

This model enables us to systematically test the direct effects of image complexity, the moderating role of readability, and the non-linear relationship between texture homogeneity and review helpfulness.

5.2. Result Analysis

This section presents the findings of the negative binomial regression analysis. We first examine the direct effects of image complexity (color diversity and texture homogeneity) on review helpfulness, followed by an analysis of the moderating role of text readability. The results are summarized in

Table 5. The results of negative binomial regression.

	(1)	(2)	(3)	(4)	(5)	(6)
Rating Score	−0.159 ***	−0.164 ***	−0.161 ***	−0.156 ***	−0.154 ***	−0.157 ***
	(0.013)	(0.013)	(0.013)	(0.013)	(0.013)	(0.013)
Sentiment Score	0.109 ***	0.107 ***	0.100 ***	0.106 ***	0.102 ***	0.103 ***
	(0.017)	(0.017)	(0.017)	(0.017)	(0.017)	(0.017)
Sentence Count	0.035	0.032 *	0.146 ***	0.034 *	0.162 ***	0.163 ***
	(0.019)	(0.019)	(0.029)	(0.019)	(0.029)	(0.029)
Image Count	0.201 ***	0.210 ***	0.201 ***	0.206 ***	0.198 ***	0.207 ***
	(0.016)	(0.016)	(0.016)	(0.016)	(0.016)	(0.016)
Word Count	0.602 ***	0.600 ***	0.494 ***	0.598 ***	0.478 ***	0.474 ***
	(0.023)	(0.023)	(0.030)	(0.023)	(0.030)	(0.030)
Review Lifespan	0.341 ***	0.327 ***	0.326 ***	0.332 ***	0.330 ***	0.325 ***
	(0.015)	(0.015)	(0.015)	(0.015)	(0.015)	(0.015)
Product Type	0.170 ***	0.201 ***	0.196 ***	0.130 ***	0.119 ***	0.115 ***
	(0.037)	(0.038)	(0.038)	(0.040)	(0.041)	(0.041)
Color Diversity		0.072 ***	0.074 ***			0.100 ***
		(0.015)	(0.015)			(0.032)
Homogeneity				0.089 ***	0.088 ***	0.122 ***
				(0.023)	(0.023)	(0.025)
Readability			0.118 ***		0.133 ***	0.120 ***
			(0.022)		(0.023)	(0.025)
Color Diversity ×			−0.005			0.051
Readability			(0.012)			(0.032)
Homogeneity2				−0.054 ***	−0.054 ***	−0.026 **
				(0.008)	(0.008)	(0.012)
Homogeneity ×					0.077 ***	0.097 ***
Readability					(0.020)	(0.024)
Homogeneity2 ×					−0.019 ***	−0.005
Readability					(0.007)	(0.011)
Observation	25,121	25,121	25,121	25,121	25,121	25,121

Standard errors in parentheses. * p < 0.1, ** p < 0.05, *** p < 0.01.

.

5.2.1. The Main Effect of Image Complexity

To test H1 and H2, we analyze the effects of color diversity and texture homogeneity on the helpfulness of reviews. Model (2) introduces color diversity, while Model (4) incorporates texture homogeneity and its quadratic term. Model (6) includes both variables along with interaction terms.

As shown in Model (2) and Model (6) in Table 5, color diversity exhibits a positive and significant effect on review helpfulness (β = 0.072, p < 0.01 in Model 2; β = 0.100, p < 0.01 in Model 6). This result aligns with Information Diagnosticity Theory, which suggests that richer visual information enhances consumers’ ability to extract valuable insights from an image [12]. Furthermore, Dual Coding Theory supports the notion that visually diverse images engage consumers more effectively by activating the non-verbal processing channel [42]. The significant positive coefficient indicates that reviews containing images with greater color variation are perceived as more helpful, likely because they provide more diagnostic visual details that aid product evaluation. This finding supports H1, confirming that color diversity is positively associated with review helpfulness.

In contrast to color diversity, texture homogeneity demonstrates a non-linear, inverted U-shaped relationship with review helpfulness. As seen in Model (4) and Model (6), texture homogeneity has a positive linear effect (β = 0.089, p < 0.01 in Model 4; β = 0.122, p < 0.01 in Model 6), but its quadratic term (Homogeneity²) is negative and significant (β = −0.054, p < 0.01 in Model 4; β = −0.026, p < 0.05 in Model 6). Figure 2 shows the inverted U-shaped relationship.

We also conducted the U-test to validate the inverse U-shaped relationship. According to the results, there is strong evidence supporting an inverse U-shaped relationship between homogeneity and vote count. The turning point of the curve is estimated at 0.826, and the 95% Fieller confidence interval for the turning point [0.272, 1.314] lies well within the observed range of homogeneity values [−2.08, 4.73]. Both slopes on either side of the turning point are statistically significant (p < 0.001), with a positive slope before the peak and a negative slope after, confirming the presence of a significant inverted U-shaped effect.

These findings indicate that review helpfulness increases as texture homogeneity rises from low to moderate levels but decreases when homogeneity becomes excessively high. This also supports Berlyne’s Aesthetic Theory, which posits that moderate complexity is preferred, whereas excessive simplicity or excessive uniformity can reduce engagement [68]. From an information Diagnosticity perspective, a moderate level of texture homogeneity may provide precise visual details, enhancing consumers’ ability to evaluate a product or service. However, when homogeneity is too high, the image may become too smooth and lack distinct visual cues, reducing its informativeness. These results confirm H2, validating the inverted U-shaped relationship between texture homogeneity and review helpfulness.

5.2.2. The Moderating Effect of Readability

According to Table 5, the interaction term Color Diversity × Readability is not statistically significant (β = −0.005, p > 0.1 in Model 3; β = 0.051, p > 0.1 in Model 6), suggesting that text readability does not significantly strengthen or weaken the relationship between color diversity and review helpfulness. This result does not support H3. One plausible explanation lies in the Visual Dominance Effect [69], which posits that consumers often rely heavily on salient visual cues during the early stages of information processing. Highly vivid and colorful images may dominate cognitive processing and reduce the influence of accompanying text. In this case, even when the textual review is easy to read, the impact of color diversity may have already occurred during the initial automatic processing phase. As such, the additive effect of readability may be limited, especially when visual salience is high. This suggests that when image color is intense or vibrant, consumers might anchor their evaluations on visual impressions alone, with limited influence from the accompanying text.

For texture homogeneity, the interaction effect Homogeneity × Readability is positive and significant (β = 0.077, p < 0.01 in Model 5; β = 0.097, p < 0.01 in Model 6), indicating that higher readability enhances the positive impact of moderate homogeneity on review helpfulness. In Model (5), the interaction term for the quadratic effect (Homogeneity² × Readability) is significant (β = −0.019, p < 0.01), indicating that readability moderates the curvilinear relationship between texture homogeneity and review helpfulness. This suggests that when text readability is high, the inverted U-shaped relationship between homogeneity and review helpfulness becomes more pronounced. Thus, H4 is supported. This pattern aligns with Information Diagnosticity Theory, which posits that consumers evaluate review content more favorably when visual and textual cues are both clear and congruent. When text readability is high, the inverted U-shaped relationship between texture homogeneity and review helpfulness becomes more pronounced. This suggests that a higher degree of alignment between what consumers read and what they see enhances information processing, thereby increasing the perceived helpfulness of the review. To examine the moderating effect of Readability, the sample was split into two groups based on the median value of Readability: a high-readability group and a low-readability group. The moderating effect of readability is illustrated in Figure 3.

However, in Model (6) where both Color Diversity × Readability and Homogeneity × Readability are included, the interaction term for the quadratic effect of Homogeneity becomes non-significant (β = −0.005, p > 0.1). This suggests that when considering the combined moderating effects of readability on both color diversity and texture homogeneity, the influence of readability on the curvature of the homogeneity–helpfulness relationship weakens. One possible explanation is that the inclusion of the color diversity interaction absorbs some of the variance previously attributed to the quadratic homogeneity interaction, leading to a reduction in statistical significance. From this point, H4 is not supported.

Following Information Diagnosticity Theory, readability facilitates information integration, particularly when visual cues are moderately complex. However, when both color diversity and texture homogeneity interact with readability, consumers may rely more on color richness for visual information processing, diminishing the moderating effect of readability on the curvilinear impact of texture homogeneity. Additionally, Dual Coding Theory suggests that when both textual and visual channels are engaged, excessive homogeneity may play a less dominant role, leading to a weaker moderating effect on its quadratic term. According to the evidence in this study, we conclude that H4 is partially supported. This suggests that while readability moderates the impact of texture homogeneity on review helpfulness, its effect on the curvature of this relationship is contingent upon whether color diversity is also considered as an interactive factor.

5.3. Robustness Test

5.3.1. Model Replacement

To ensure the robustness of our findings, we conduct a model replacement test by employing two alternative regression approaches: Zero-Inflated Negative Binomial (ZINB) regression and logistic regression (Logit). Given that the dependent variable contains a substantial number of zero values, ZINB is an appropriate model to account for excess zeros and overdispersion [67]. Additionally, we transform the continuous variable Vote Count into a binary one (1 if the review received at least one vote, and 0 otherwise), and apply a logit model to examine the robustness of our results in a different analytical framework. The results of these analyses are presented in

Table 6. Results of zero-inflated negative binomial (ZINB) and logit regression.

	(1)	(2)	(3)	(4)	(5)	(6)
	ZINB	Logit	ZINB	Logit	ZINB	Logit
Rating Score	−0.161 ***	−0.160 ***	−0.154 ***	−0.157 ***	−0.157 ***	−0.157 ***
	(0.013)	(0.017)	(0.013)	(0.017)	(0.013)	(0.017)
Sentiment Score	0.100 ***	0.035 *	0.102 ***	0.035 **	0.103 ***	0.036 **
	(0.017)	(0.018)	(0.017)	(0.018)	(0.017)	(0.018)
Sentence Count	0.146 ***	0.201 ***	0.162 ***	0.207 ***	0.163 ***	0.207 ***
	(0.029)	(0.031)	(0.029)	(0.031)	(0.029)	(0.031)
Image Count	0.201 ***	0.245 ***	0.198 ***	0.244 ***	0.207 ***	0.249 ***
	(0.016)	(0.017)	(0.016)	(0.017)	(0.016)	(0.017)
Word Count	0.494 ***	0.284 ***	0.478 ***	0.278 ***	0.474 ***	0.276 ***
	(0.030)	(0.032)	(0.030)	(0.032)	(0.030)	(0.032)
Review Lifespan	0.326 ***	0.353 ***	0.330 ***	0.355 ***	0.325 ***	0.353 ***
	(0.015)	(0.016)	(0.015)	(0.016)	(0.015)	(0.016)
Product Type	0.196 ***	0.292 ***	0.119 ***	0.223 ***	0.115 ***	0.221 ***
	(0.038)	(0.040)	(0.041)	(0.043)	(0.041)	(0.043)
Color Diversity	0.074 ***	0.046 ***			0.100 ***	0.055 *
	(0.015)	(0.016)			(0.032)	(0.032)
Homogeneity			0.088 ***	0.082 ***	0.122 ***	0.101 ***
			(0.023)	(0.024)	(0.025)	(0.026)
Readability	0.118 ***	0.164 ***	0.132 ***	0.176 ***	0.120 ***	0.163 ***
	(0.022)	(0.024)	(0.023)	(0.025)	(0.025)	(0.026)
Color Diversity ×	−0.005	0.012			0.050	0.048
Readability	(0.012)	(0.014)			(0.032)	(0.031)
Homogeneity2			−0.054 ***	−0.045 ***	−0.026 **	−0.029 **
			(0.008)	(0.009)	(0.012)	(0.013)
Homogeneity ×			0.077 ***	0.035 *	0.097 ***	0.053 **
Readability			(0.020)	(0.021)	(0.024)	(0.024)
Homogeneity2 ×			−0.019 ***	−0.014 *	−0.005	−0.000
Readability			(0.007)	(0.009)	(0.011)	(0.012)
Observation	25,121	25,121	25,121	25,121	25,121	25,121

Standard errors in parentheses. * p < 0.1, ** p < 0.05, *** p < 0.01.

.

The results from both ZINB models and Logit models remain largely consistent with those obtained from the negative binomial regression. Specifically, color diversity consistently exhibits a positive relationship with review helpfulness across models. Texture homogeneity demonstrates a robust inverted U-shaped relationship, reinforcing the notion that moderate texture complexity optimizes review helpfulness. For the moderating effect, text readability enhances the impact of texture homogeneity, but its moderating role on the quadratic term is less consistent when multiple interaction effects are included. These results strengthen the validity of our conclusions and indicate that the observed relationships are not driven by the specific choice of statistical model.

5.3.2. Winsorization of Independent Variables

To further verify the robustness of our findings and test whether the results are sensitive to extreme values, we perform Winsorization on the key independent variables and re-estimate the model using negative binomial regression [70]. Specifically, given that Color Diversity is left-skewed, we apply a 1% Winsorization on the left tail. Conversely, as Homogeneity is right-skewed, we perform a 1% Winsorization on the right tail. This process mitigates the influence of outliers while preserving the overall distributional characteristics of the data. The results of this analysis are presented in

Table 7. The results after Winsorization of the independent variables.

	(1)	(2)	(3)
Rating Score	−0.161 ***	−0.154 ***	−0.157 ***
	(0.013)	(0.013)	(0.013)
Sentiment Score	0.100 ***	0.102 ***	0.103 ***
	(0.017)	(0.017)	(0.017)
Sentence Count	0.146 ***	0.163 ***	0.164 ***
	(0.029)	(0.029)	(0.029)
Image Count	0.201 ***	0.198 ***	0.206 ***
	(0.016)	(0.016)	(0.016)
Word Count	0.494 ***	0.477 ***	0.473 ***
	(0.030)	(0.030)	(0.030)
Review Lifespan	0.325 ***	0.329 ***	0.324 ***
	(0.015)	(0.015)	(0.015)
Product Type	0.198 ***	0.117 ***	0.114 ***
	(0.038)	(0.041)	(0.041)
Color Diversity	0.076 ***		0.091 ***
	(0.015)		(0.031)
Homogeneity		0.093 ***	0.123 ***
		(0.022)	(0.025)
Readability	0.119 ***	0.134 ***	0.120 ***
	(0.022)	(0.024)	(0.025)
Color Diversity ×	−0.005		0.050
Readability	(0.012)		(0.031)
Homogeneity2		−0.059 ***	−0.034 ***
		(0.009)	(0.012)
Homogeneity ×		0.078 ***	0.098 ***
Readability		(0.020)	(0.023)
Homogeneity2 ×		−0.020 ***	−0.006
Readability		(0.008)	(0.012)
Observation	25,121	25,121	25,121

Standard errors in parentheses. *** p < 0.01.

. The Winsorization-adjusted regression results remain consistent with our main findings, indicating that the observed relationships are not driven by extreme values or outliers, further enhancing the validity and reliability of our conclusions.

5.4. Heterogeneity Analysis

To further examine whether the relationship between image complexity and review helpfulness varies across reviews of different product categories, we conduct a heterogeneity analysis by separately estimating the regression model for hotel reviews and travel reviews. This enables us to examine potential differences in how visual and textual cues affect consumer perceptions of reviews in various contexts.

Table 8. The results of the hotels and travel dataset.

	Hotel			Travel
	(1)	(2)	(3)	(4)	(5)	(6)
Rating Score	−0.196 ***	−0.191 ***	−0.193 ***	−0.012	−0.013	−0.006
	(0.018)	(0.017)	(0.018)	(0.019)	(0.019)	(0.019)
Sentiment Score	0.001	−0.004	−0.005	0.115 ***	0.117 ***	0.116 ***
	(0.026)	(0.026)	(0.026)	(0.023)	(0.023)	(0.023)
Sentence Count	0.157 ***	0.165 ***	0.164 ***	0.230 ***	0.231 ***	0.237 ***
	(0.043)	(0.043)	(0.044)	(0.041)	(0.042)	(0.041)
Image Count	0.288 ***	0.285 ***	0.287 ***	0.107 ***	0.101 ***	0.115 ***
	(0.023)	(0.023)	(0.023)	(0.021)	(0.021)	(0.021)
Word Count	0.074 *	0.064	0.066	0.642 ***	0.646 ***	0.635 ***
	(0.042)	(0.043)	(0.043)	(0.041)	(0.041)	(0.041)
Review Lifespan	0.313 ***	0.315 ***	0.314 ***	0.343 ***	0.351 ***	0.340 ***
	(0.022)	(0.022)	(0.022)	(0.020)	(0.020)	(0.020)
Color Diversity	−0.009		0.019	0.091 ***		0.225 ***
	(0.023)		(0.026)	(0.021)		(0.056)
Homogeneity		0.126 ***	0.132 ***		−0.005	0.098 **
		(0.026)	(0.028)		(0.029)	(0.039)
Readability	0.142 ***	0.145 ***	0.143 ***	0.190 ***	0.227 ***	0.231 ***
	(0.039)	(0.040)	(0.040)	(0.031)	(0.033)	(0.036)
Color Diversity ×	0.011		0.024	−0.002		0.018
Readability	(0.020)		(0.022)	(0.017)		(0.058)
Homogeneity2		−0.035 ***	−0.033 **		−0.028 **	0.027
		(0.013)	(0.013)		(0.013)	(0.019)
Homogeneity ×		0.010	0.019		0.100 ***	0.116 ***
Readability		(0.027)	(0.028)		(0.030)	(0.042)
Homogeneity2 ×		0.004	0.005		−0.042 ***	−0.039 **
Readability		(0.011)	(0.011)		(0.012)	(0.018)
Observation	14,111	14,111	14,111	11,010	11,010	11,010

Standard errors in parentheses. * p < 0.1, ** p < 0.05, *** p < 0.01.

displays the corresponding analysis results.

5.4.1. Differences in the Effect of Image Complexity

The influence of image complexity on review helpfulness varies significantly between hotel and travel review datasets. For color diversity, the results show no significant effect in hotel reviews (β = −0.009, p > 0.1 in Model 1; β = 0.019, p > 0.1 in Model 3), suggesting that visual richness in color does not meaningfully impact perceived helpfulness in this context. In contrast, color diversity demonstrates a strong positive effect in travel reviews (β = 0.091, p < 0.01 in Model 4; β = 0.225, p < 0.01 in Model 6), indicating that vibrant, colorful images enhance the perceived informativeness of travel-related content. This aligns with Information Diagnosticity Theory, as colorful visuals may convey richer sensory and contextual cues, aiding consumers in evaluating experiential attributes of destinations.

For texture homogeneity, the findings also differ. In hotel reviews, texture homogeneity shows a significant positive effect (β = 0.126, p < 0.01 in Model 2; β = 0.132, p < 0.01 in Model 3), and the quadratic term is negative and significant (β = −0.035, p < 0.01 in Model 2; β = −0.033, p < 0.05 in Model 3), confirming an inverted U-shaped relationship. This suggests that moderately homogeneous textures, which balance clarity and detail, are most effective in enhancing perceived helpfulness. For travel reviews, however, the linear effect is only significant in Model 6 (β = 0.098, p < 0.05), and the quadratic term is not significant, indicating that the inverted U-shaped pattern does not hold as strongly.

These differences reflect the distinct goal orientations associated with hedonic versus utilitarian consumption. Travel products are typically hedonic in nature [71], where aesthetic experiences, emotional immersion, and atmospheric cues play a central role in consumers’ evaluations [72]. In this context, Color-rich images may evoke affective responses and signal scenic quality or cultural richness, which aligns with experiential consumption goals. Conversely, hotel or homestay decisions are more utilitarian, where consumers rely more heavily on clear visual cues that convey factual, material details [71,73], typically encoded in texture and structure. Thus, texture-related image complexity becomes particularly salient for utilitarian evaluations, such as accommodation choices, while color-related complexity holds greater relevance for hedonic evaluations, such as travel experiences. These findings extend the literature by demonstrating that the diagnostic value of visual complexity is not universal but somewhat contingent upon the nature of the consumption context.

5.4.2. Differences in the Moderating Effect of Readability

The moderating role of text readability in the relationship between texture homogeneity and review helpfulness reveals notable asymmetries across product categories. In hotel reviews, neither the interaction term Homogeneity × Readability (β = 0.010, p > 0.1 in Model 2; β = 0.019, p > 0.1 in Model 3) nor the quadratic interaction Homogeneity² × Readability (β = 0.004, p > 0.1 in Model 2; β = 0.005, p > 0.1 in Model 3) reaches statistical significance. This suggests that, within utilitarian settings, such as accommodations, the effect of texture homogeneity on perceived helpfulness is robust and relatively unaffected by the clarity of the textual content. By contrast, in travel reviews, a domain characterized by hedonic and experiential consumption, the main effect of texture homogeneity is only marginally significant (β = −0.005, p > 0.1 in Model 4; β = 0.098, p < 0.05 in Model 6), indicating that on its own, texture does not strongly predict perceived helpfulness. However, when moderated by text readability, both the linear (β = 0.100, p < 0.01 in Model 5; β = 0.116, p < 0.01 in Model 6) and quadratic interaction terms (β = −0.042, p < 0.01 in Model 5; β = −0.039, p < 0.05 in Model 6) become significant.

This pattern suggests that text readability plays a compensatory and enabling role in the context of hedonic consumption. In the absence of highly diagnostic image textures, readable text may guide the consumer to more effectively interpret the visual cues, thereby amplifying the perceived informativeness of moderately textured images. However, when texture homogeneity becomes excessive, even high readability cannot overcome the perceptual limitation because clear text only further highlights the absence of meaningful visual details. This finding also aligns with Information Diagnosticity Theory, which points out that the effectiveness of information arises from alignment between the form of information. In travel reviews, when text is easy to process, consumers may become more sensitive to the alignment or misalignment between descriptive text and visual representation. Thus, the readability of text not only enhances the usefulness of moderate texture complexity but also amplifies the adverse effect of overly uniform, visually non-diagnostic imagery. These findings contribute to a deeper understanding of how product categories shape consumer engagement with visual and textual review cues.

6. Discussion

6.1. Theoretical Implications

This study advances the literature on online review effectiveness by integrating established frameworks with recent empirical insights. Our findings extend Information Diagnosticity Theory [12] by demonstrating that the quality and richness of visual cues, which are color diversity and texture homogeneity, play a critical role in reducing consumer uncertainty [36]. Recent evidence by Zhao et al. [74] corroborates this view, showing that vibrant images in social networks provide detailed and contextually relevant information, thereby enhancing customer engagement.

Furthermore, our research refines Dual Coding Theory [42,63,75] by highlighting the importance of multimodal integration in online information processing. While classic theory emphasizes the additive benefits of processing both verbal and non-verbal information, our results indicate that text readability significantly moderates the effectiveness of visual cues. Eitel and Scheiter [61] demonstrated that presenting clear and concise textual information alongside images facilitates better understanding. Our analysis further clarifies the differentiated role of text readability across product categories. For hotel consumers, the texture and clarity of images directly convey functional value, rendering the influence of text readability negligible. In contrast, for travelers, moderately textured images paired with readable text amplify the vividness and emotional tone of the content, thereby enhancing the perception of hedonic value.

Additionally, the inverted U-shaped relationship we observe between texture homogeneity and review helpfulness—especially in hotel reviews—offers an extension of Berlyne’s Aesthetic Theory [68]. Although earlier studies established that moderate complexity optimizes aesthetic appeal, our findings reveal that this optimal balance varies by product category. In hotel consumption contexts, where utilitarian value is prioritized, image clarity plays a more critical role. Moderate texture homogeneity enhances the perceived quality of reviews, suggesting that aesthetic balance is a key determinant in consumers’ evaluative judgments for utilitarian products.

By bridging classic theories, our research deepens the theoretical understanding of multimodal information processing in digital environments. These contributions offer a new perspective for future research to investigate dynamic interactions among sensory channels in various product contexts.

6.2. Managerial Implications

Our findings offer actionable insights for online review platforms and businesses aiming to optimize the effectiveness of user-generated content. First, for hedonic products such as travel experiences, color diversity has a positive impact on review helpfulness. Platforms should encourage users to upload vibrant, context-rich images when reviewing destinations. Prior research suggests that visually engaging content improves consumer engagement [74]. Platforms can implement features, such as AI-driven image recommendations or review prompts, to help users select the most relevant images. Besides, for utilitarian products, such as hotel accommodations, moderate texture homogeneity enhances the informativeness of reviews. Platforms can provide automated image enhancement tools that balance sharpness and detail to optimize image quality for hotel reviews. This finding aligns with those of Ma et al. and Zhang et al. [76,77], who emphasize that visual clarity improves consumer trust in online content.

Secondly, given the hedonic attribute of travel, text readability significantly amplifies the effect of image complexity in travel-related reviews. Platforms should incorporate readability assessments or AI-powered writing assistance tools to help users create clearer and structured reviews. Prior research suggests that cognitive load is reduced when textual and visual elements are well-integrated [10,75]. Review templates or suggested sentence structures may further assist users in composing easily digestible content.

Our findings suggest that review helpfulness is driven by different factors depending on the product category. In travel reviews that emphasize hedonic experiences, color diversity and text readability are crucial, whereas in hotel reviews that prioritize utilitarian experiences, texture homogeneity plays a more significant role. By implementing these strategies, businesses and review platforms can enhance user engagement, improve the informativeness of reviews, and ultimately influence consumer purchase decisions, contributing to a more effective digital consumer experience.

6.3. Limitations and Future Research

While this study offers valuable insights into the impact of image complexity on the perceived helpfulness of online reviews, several limitations exist.

First, the data used in this research are limited to two specific product categories, namely hotels and travel, collected from two Chinese online travel platforms: Ctrip and Tongcheng. This narrow focus may limit the generalizability of the findings to other product types or cultural contexts. Future research could expand the scope by examining a broader range of products and services across different e-commerce platforms and in various cultural settings. This would help determine whether the observed effects of color diversity and texture homogeneity are consistent across different consumer markets and product categories.

Second, this study relies on computational measures of image complexity, specifically color diversity and texture homogeneity. While these metrics effectively capture key visual properties, they do not account for contextual relevance or semantic meaning within images. Consumers may perceive an image as helpful not solely based on its complexity but also on its alignment with the review content and its ability to depict key product features. Future research could integrate advanced computer vision techniques, such as object recognition, scene analysis, or sentiment-based image classification, to assess how the content and meaning of images influence review helpfulness.

Third, this study focuses on readability as a textual feature, but other linguistic factors, such as argument quality, writing style, sentiment polarity, and emotional appeal, may also moderate the effect of visual cues. For instance, highly emotional language may increase engagement, while concise, well-structured arguments may enhance credibility. Future research could leverage natural language processing (NLP) techniques to explore how different textual characteristics interact with visual elements in shaping consumer perceptions of review helpfulness.

Finally, this study utilizes observational data from online reviews, which limits the ability to make causal inferences. Although robust regression models and robustness checks were employed, the study cannot fully isolate the causal relationships between image complexity, readability, and review helpfulness. Future research could employ experimental designs to manipulate image characteristics and textual readability in controlled settings, providing stronger causal evidence on how multimodal information processing influences consumer decision-making.

By addressing these limitations, future research can deepen our understanding of how consumers integrate visual and textual information in online reviews, further advancing theories in consumer behavior, online marketing, and human-information interaction.