Topic Modeling of Positive and Negative Reviews of Soulslike Video Games

Abstract

Soulslike games are renowned for their challenging gameplay and distinctive design. To examine player reception of this genre, 993,932 user reviews of 21 Soulslike video games were collected from the Steam platform, of which 418,483 were tagged as English and analyzed. Latent Dirichlet Allocation (LDA) was applied to identify and compare thematic patterns across positive and negative reviews. The resulting topics were grouped into five categories: aesthetics, gameplay mechanics, feelings, bugs/issues, and miscellaneous. Positive reviews emphasized aesthetics and atmosphere, whereas negative reviews focused on gameplay mechanics and technical issues. Notably, emotional tone differed significantly between review types. Overall, these results may benefit game developers refining design elements, researchers investigating player experience, and critics analyzing the reception of Soulslike games. Furthermore, the study provides a basis for understanding player perspectives in Soulslike games and establishes a foundation for comparative research with newer titles such as Elden Ring.

1. Introduction

The first game in the Souls franchise was released in 2009 for the PlayStation 3 (PS3). This was titled Demon’s Souls and was developed by the Japanese game studio FromSoftware, directed by Hidetaka Miyazaki. It received a mixed reception at launch but ultimately became a commercial success. Consequently, three sequels were greenlit and developed by the same studio: Dark Souls I, II, and III, released in 2011, 2014, and 2016, respectively. Unlike Demon’s Souls, the Dark Souls games were also released on PC. For this reason, the first Dark Souls is often regarded as the game that established the franchise [1,2].

The term “Soulslike” is frequently used to describe games with key characteristics resembling those of the Dark Souls games. These games share several defining traits [3]. According to Jagoda, they present clear endpoints or goals for the players but place multiple challenges in the players’ path to prevent them from reaching those goals [4]. Thus, one of the key features of these games is their demanding gameplay. They feature a steep difficulty curve and require precise combat execution, which in turn demands a deep understanding of the game’s mechanics. Due to the punishing difficulty, player characters tend to die frequently. Therefore, players must learn from their mistakes and improve their skills to overcome enemies and environmental obstacles. This process is referred to as “pain and loss,” and it can improve players’ skills if they persevere [5], potentially increasing motivation [6,7] and creating a challenging yet rewarding flow of gameplay [8,9]. These games can therefore provide intense and satisfying gaming experiences [10], which are among the most valued by players [11].

In addition to their difficulty, these games are known for their intricate, interconnected level design [12]. Players navigate complex environments that often include hidden paths and shortcuts, which make exploration an integral part of the gameplay experience [13]. Moreover, storytelling can be embedded directly into the game’s mechanics [14]. This is the case in Soulslike games, which rely heavily on environmental and contextual storytelling [15,16,17]. Rather than relying on explicit exposition or cutscenes, lore is conveyed through item descriptions, environmental cues, and interactions with non-playable characters. This encourages players to piece together the story themselves [18], and it is possible to complete these games without fully understanding the entire narrative. This narrative style has attracted a passionate community of players who engage in discussions, theories, and interpretations about the story.

With these gameplay mechanics and immersive worlds, Soulslike games strongly influence player experience and have left a lasting mark on the gaming industry [19,20,21,22,23,24,25,26,27]. As these games gained popularity, the Soulslike genre emerged, with other developers adopting similar design philosophies in their own titles over the years [1,2,28]. Over time, the Soulslike genre has evolved from a niche category into a significant trend in digital games culture [28]. Recent titles have experimented with hybridization, incorporating elements from open-world, hack-and-slash, or role-playing game systems to appeal to larger audiences while retaining the punishing core design. This transformation is mainly evident in titles such as Elden Ring, which has become the most commercially and critically successful Souls game to date. While Elden Ring is not included in this study due to its release date, it represents an important turning point in how these mechanics are understood and accepted by mainstream audiences.

It is possible to analyze player experiences and sentiments expressed in game reviews, as players often write them down [29,30]. Writing reviews also provides players with a platform to share concerns, criticisms, and suggestions for improvement [31]. Most reviews are written shortly after a game’s release, with frequency decreasing over time [32]. Common topics in game reviews include accessories, achievements, bugs, general experience, graphics, social influence, and interaction [33]. According to Khalid, several categories of complaints can be found in reviews [34]. Vasa et al. concluded that negative reviews are generally longer than positive ones [35]. Reviews can also reveal a game’s playability and popularity [36,37], although players are often more tolerant of early access games, which can be played while still in development [38]. Naturally, review topics depend on both the game and the reviewer [39].

Reviews can also influence potential buyers [40], functioning as product reviews [41]. They can reveal sentiments [42,43] and even indicate whether the reviews themselves are considered helpful [44,45,46]. By understanding game reviews, researchers and developers can gain valuable insights into a title’s strengths and weaknesses [47]. Addressing these can lead to improvements and ultimately enhance the quality of future releases [48,49].

As thousands of reviews are written daily [50], researchers have access to vast amounts of data. However, manually analyzing such large datasets is time-consuming. Topic modeling offers a valuable solution in this context. Applying these techniques to game reviews enables the automatic uncovering of latent topics, providing deeper insights into player sentiments, preferences, and concerns. Topic modeling algorithms, such as Latent Dirichlet Allocation (LDA), can identify topics within text documents [51]. LDA is a generative probabilistic model that assumes each document is a mixture of potential topics and that each topic is a distribution of words from the vocabulary. Extracting these topics helps highlight important aspects discussed by players. LDA has been applied to reviews in a limited number of studies: Wang et al. proposed a framework for detecting spam in reviews [52], while Yu et al. assessed topics in games [53]. Moreover, topic words were clustered to identify aspects in Amazon reviews [54]. Additionally, Yu et al. used LDA to compare topics in the game reviews of Dark Souls: Remastered and Dark Souls III [55]. However, these studies were limited to only two games in the genre.

The present study addresses this gap by examining a larger selection of 21 Soulslike titles available on the Steam platform. The aim is to identify thematic structures in positive and negative reviews as well as to understand how players construct feedback about these games. Since these reviews were gathered before the release of Elden Ring, the results offer a pre-Elden Ring snapshot of player sentiment and provide a foundation for future comparative analyses involving newer titles. Accordingly, the research questions (RQs) are as follows:

• RQ1: What are the themes and topics discussed in positive and negative reviews of Soulslike games?
• RQ2: Are there any significant differences in topics between positive and negative reviews of Soulslike games?

The paper is structured as follows. Section 2 describes the materials and methods used in the study, from review scraping to data analysis. Section 3 details the results. The discussion is presented in Section 4, and the conclusions are presented in Section 5.

2. Materials and Methods

This section is split into three subsections. The first one presents the chosen digital game distribution platform from which the reviews were scraped. The scraping process is detailed in the second subsection, and the third subsection presents the data analysis.

2.1. The Steam Digital Game Distribution Platform

According to Lin et al., Steam was the largest digital game distribution platform in 2017 [31]. As of March 2021, it remained one of the largest, hosting more than 50,000 video games [56]. Players can post reviews on each game’s page, which is considered one of Steam’s most important features. Reading reviews is free, and anyone can browse them. This accessibility is one of the reasons Steam was chosen for this study.

On Steam, every video game has its own store page and the reviews for each game appear on a dedicated subpage. While a review contains multiple fields of information [57], only the following were used in this study: the game’s name, the language of the review, the textual content, and whether the game was recommended. Steam does not have a numerical rating system; instead, players can give a game either a “Recommended” or “Not Recommended” rating, which correspond to a positive or negative review, respectively. In addition to the recommendation choice, reviews include a textual component in which players can describe their experiences. The development team behind Steam, Valve Corporation, provides an application programming interface (API) that allows reviews to be scraped [58].

2.2. The Scraping Process

Before scraping began, the games themselves had to be selected. Using Steam’s tag system, 21 Soulslike games were identified. However, because the tags are user-generated, each game required careful evaluation. For example, the “Soulslike” tag sometimes appeared on unrelated titles, including video editing software and games that did not fit the genre. To ensure consistency and relevance, the following selection criteria were applied:

• High difficulty and frequent player character death as progression feedback;
• Deliberate, timing-based combat involving dodging, parrying, or stamina management;
• Interconnected level design that encourages exploration;
• Narrative conveyed through environmental storytelling rather than direct exposition.

Games lacking these mechanics or incorrectly tagged were excluded. Additionally, very recent titles with few reviews and early-access games were omitted to ensure good topic modeling results. All 21 selected games were verified through direct gameplay observation and review content analysis to confirm adherence to the genre traits above. While the Soulslike genre is difficult to define due to hybridization with other styles (e.g., Metroidvania, action role-playing game, and hack-and-slash), the selected games all center on overcoming adversity through mechanical learning and gradual mastery [28,59].

Following the selection of games, the steam_reviews Python package (version 0.1.2) was used to scrape all of their reviews in 2021 into a JSON format [60]. The package used is available under the MIT license. Overall, 993,932 reviews were scraped from the Steam platform. The jsonlite package was used to import these files into the statistical software R (version 4.3.2) for the analysis. Figure 1 shows the scraped reviews for each Soulslike video game grouped by whether reviews were positive or negative. It should be noted that this dataset was also used in prior Soulslike studies [1,2].  

2.3. Data Analysis

After scraping and importing into R, the reviews were grouped by their type. Out of the 993,932 reviews, 904,005 (90.96%) were positive, while 89,927 (9.54%) were negative. Afterward, a subset of English language reviews was created. Thus, the number of investigated reviews decreased to 418,483. Out of these, 377,334 (90.16%) were positive and 41,149 (9.84%) were negative. As can be observed, the weight of positive and negative English reviews is approximately similar to the full dataset. When the English reviews subset was created, the number of words was analyzed. This was done with the help of the stringr package. By using the str_count() function with the [\\w\’]+ regular expression, the number of words was counted in each review. To visualize the number and distribution of words, the lvplot package was used.

Before topic modeling, a text preprocessing pipeline was applied using the quanteda and tm packages:

• The dataset was restricted to reviews tagged as English by Steam metadata.
• A corpus was created with the tm package’s corpus() function using its default parameters.
• A document-feature matrix (DFM) was generated using quanteda::dfm() with remove_punct = TRUE and remove = stopwords(“english”) to remove punctuation and stopwords, respectively.
• Tokens shorter than two characters were removed with min_nchar = 2 in dfm_remove().
• Rare terms appearing in fewer than five documents were removed using min_docfreq = 5 in dfm_trim().

To assess the topics in the reviews, the lda, ldatuning, quanteda, topicmodels, and tm packages were used in R. To determine the optimal number of topics, four established methods were used [61,62,63,64], first on all reviews, then separately on positive and negative reviews. In all cases, the optimal number was 50 topics. This can be observed in Figure 2, Figure 3 and Figure 4.

Following the computation of the optimal number of topics, three LDA models were estimated using Gibbs sampling: one for all reviews, one for positive reviews, and one for negative reviews. They were estimated using Gibbs sampling with alpha = 0.1. A fixed random seed ensured reproducibility. After the models were created, they were plotted with the help of the LDAvis package. To visualize the probabilities of the top 10 words in each topic, the tidytext, ggplot2, and dplyr packages were used.

Following the estimation of the topic models, each topic was assigned to one of five thematic groups: aesthetics, gameplay mechanics, feelings, bugs/issues, and other. This classification was based on the semantic coherence of each topic’s most probable terms and their interpretability in the context of game review analysis. For example, topics dominated by terms related to combat mechanics, control schemes, or progression systems were classified under gameplay mechanics. Topics emphasizing visual design, environmental details, and sound were placed in aesthetics. Topics containing affective descriptors (e.g., “frustrating” or “rewarding”) were assigned to feelings. Terms describing bugs, glitches, or crashes were grouped as bugs/issues, while all remaining topics were placed into the other category.

The statistical distance regarding words between topics was also assessed. In the first step, the posterior probabilities were calculated for each topic. Then, using the Jensen–Shannon Divergence (JSD) from the philentropy package, statistical distances were calculated. JSD is based on the Kullback–Leibler Divergence method, and it measures the similarity between two probability distributions [65]. After the similarities were calculated, heatmaps were created using the gplots package and its heatmap.2() function to illustrate them. In all figures in this study, profane words are censored.

3. Results

This section is divided into four subsections. Section 3.1 presents the descriptive statistics. Section 3.2 and Section 3.3 describe the topics found in positive and negative reviews, respectively. Section 3.4 compares these topics.

3.1. General Results

Before examining the topics, the number of words per review was analyzed. Results show that positive reviews generally contain fewer words than negative reviews. Descriptive statistics for both review types are shown in

Table 1. Descriptive statistics of negative and positive English reviews.

	Min	Max	Median	Mean	Standard Deviation
Negative review	0.00	1885.00	21.00	75.88	152.53
Positive review	0.00	2289.00	7.00	35.88	97.63

, and their distribution is illustrated in Figure 5 using letter-value plots. This type of plot is designed for large datasets and provides better insight into distribution tails beyond the quartiles [66].

These results are consistent with prior results in broader Steam review studies [31], although the magnitude is greater in Soulslike reviews. In this dataset, positive reviews had a median length of 7 words, while negative reviews had a median of 21 words. This suggests that negative experiences in this genre prompt players to elaborate more, possibly to justify a “Not Recommended” rating in a niche genre known for its difficulty.

Regression analysis was also performed to check whether the type of review affects its length. The results of the examination show that review type has a significant effect of review length.

Table 2. The effect of review type on word numbers.

	Estimate	Standard Error	t Value	Pr(>|t|)
Intercept	110.35	0.61	179.69	<0.001
Positive review	−60.12	0.64	−92.96	<0.001

shows the strength of the effect. As can be observed, negative review type was chosen as a basis variable. According to the results presented in Table 2, the length of positive reviews is significantly shorter.

Next, topic size, intertopic distances, and the most salient words were analyzed (Figure 6). Topic size refers to the number of reviews assigned to a topic, while distance indicates topical similarity. As expected, the first topic in each model is the largest, and the last is the smallest.

Figure 6 shows that several topics are related to one another. Furthermore, based on the top 30 most salient words, several terms occur across multiple topics. “Game” is the most frequent word, whereas “died” is the least frequent. Notably, both positive and negative terms appear, along with words related to Soulslike game mechanics and storytelling. To gain a clearer understanding of the topics, the probabilities of the top 10 words in each topic were examined. The results of this analysis are shown in Figure 7.

Judging from the salient terms and the probabilities of the top 10 words in each topic, topics can be categorized into five groups. Topics that contain terms about various aspects of Soulslike games (group 1); topics that contain terms about gameplay mechanics (group 2); topics that contain terms about the feelings of players (group 3), topics that report issues or bugs (group 4), and other topics (group 5). As shown in

Table 3. Groups and their corresponding topics in the general results.

Group	Topics
1	4, 11, 13, 14, 26, 27, 28, 29, 32, 37, 41, 43, 44, 46
2	1, 2, 6, 8, 9, 15, 16, 33, 36, 38, 39, 42, 45
3	3, 5, 10, 12, 17, 19, 20, 21, 22, 23, 24, 25, 30, 31, 34, 35, 40, 49, 50
4	47, 48
5	7, 18

, the following topics can be placed in these groups.

The word probabilities in topics were also assessed using JSD. The results can be observed in Figure 8.

As can be seen in Figure 8, statistical distances are closer between word probabilities in topics in the lower left part of the heatmap. These include topics 41, 46, 22, 21, 29, 48, 1, 38, 11, 14, 15, 24, 36, 33, 9, 16, 27, 26, and 39. Therefore, the words in these topics have similar probabilities of appearing in their respective topics. When compared to the remaining topics, the distances become larger.

3.2. Topics in Positive Reviews

Next, the intertopic distance and the salient words in positive reviews were investigated. The results are shown in Figure 9.

The intertopic distance map for positive reviews (Figure 9) closely resembles that of all reviews. Similarity can also be observed in the case of the top 30 salient words. Almost all are the same, with the exception of the words “worth” and “buy”. When all reviews were investigated, the words “just” and “pretty” could be observed in their places. As in the case of all reviews, both positive and negative terms can be seen among the salient ones. Like before, the probabilities of the top 10 words in each topic were analyzed. The results of this analysis can be observed in Figure 10.

The words in positive reviews were grouped into the same groups as all reviews were. The results are the following, as shown in

Table 4. Groups and their corresponding topics in positive reviews.

Group	Topics
1	1, 8, 9, 11, 18, 20, 21, 22, 23, 26, 27, 30, 31, 32, 35, 36, 39, 41, 42, 45, 46
2	5, 6, 10, 13, 44, 48
3	2, 3, 7, 12, 14, 15, 16, 19, 24, 28, 29, 38, 40, 43, 47, 49, 50
4	17, 25, 37
5	4, 33

.

This classification shows that positive reviews more often emphasize the core aesthetic and experiential elements of Soulslike games (Group 1) than gameplay complaints or issues. The presence of Group 4 topics (bugs/issues) in positive reviews indicates that even satisfied players may mention technical shortcomings, though they do not affect the overall recommendation.

Next, to understand the differences between word probabilities in various topics, they were investigated using JSD. The results can be seen in Figure 11.

According to Figure 11, topics 37, 33, and 34 were the closest to each other regarding word probabilities. The distance between word probabilities increases with each comparison. It can also be noted that topic 38 is the most dissimilar to the others and forms the tallest branch in the dendrogram. This indicates that this topic has the most different word probabilities from the other topics.

3.3. Topics in Negative Reviews

Next, the negative reviews were investigated. Similarly, the intertopic distances and the salient terms were assessed. The results are shown in Figure 12.

As observed in Figure 12, many topics are far from each other, and there are only a few topics that are conjoined. The top 30 salient words are vastly different from those of all and positive reviews. The newly appearing salient terms are the following: “PC”, “roll”, “enemies”, “controller”, “bosses”, “money”, “port”, “controls”, “time”, “buy”, “first”, “worst”, “attack”, “design”, and the s-word as well as the f-word. This lexical shift reflects a different tone in negative reviews, which tend to emphasize frustration with specific gameplay mechanics (e.g., dodging or “roll”), technical performance (“port”, “PC”), and financial regret (“money”, “buy”). Profanity appears more frequently, consistent with the stronger emotional expression seen in prior sentiment analyses of negative reviews.

The next step was to better understand the context of words in negative reviews. Therefore, the probabilities of the top 10 words were assessed in each topic. The results of this investigation can be seen in Figure 13.

The words in negative reviews were also grouped into the same groups as previously. The results are shown in

Table 5. Groups and their corresponding topics in positive reviews.

Group	Topics
1	1, 3, 4, 7, 19, 20, 31, 32, 35, 36, 37, 39, 44
2	6, 10, 11, 12, 14, 21, 23, 24, 25, 26, 27, 42, 43, 45, 48, 49
3	2, 5, 8, 9, 17, 22, 29, 33, 34, 38, 40, 41, 46, 47, 50
4	15, 16, 18, 28
5	13, 30

.

Compared to positive reviews, the number of topics that are related to gameplay complaints is higher (Group 2). This reinforces the idea that dissatisfaction is more varied and mechanic-specific than praise, which is typically more unified around atmosphere and design. The topics in Group 4 that address bugs and performance issues are more frequent and include several mentions of accessibility limitations (e.g., controller setup and lack of difficulty settings).

Next, the differences between word probabilities in the various topics were assessed using JSD. The results of the analysis can be observed in Figure 14.

Figure 14 is similar to Figure 8 in that the lower-left corner contains the topics that are closest in probabilistic terms. Keep in mind that the values are much smaller as well. In this regard, the probabilities of words in negative reviews can be considered similar to each other between topics.

3.4. Comparison of Topics Between Review Types

As mentioned in Figure 9 and Figure 12, the intertopic distance map is vastly different between the two review types. Several topics in positive reviews are close to each other, and most of them are even conjoined. The case is different when the intertopic distance map of negative reviews is analyzed. This indicates that the topics found in positive reviews are more similar to each other than those observed in negative reviews. Also, while positive reviews frequently focus on a shared core set of features such as atmosphere, design, and overall satisfaction, negative reviews reflect more individualized or situational frustrations.

The last step was to compare the probabilities of words in both types of reviews. The results of the comparison can be observed in Figure 15.

Probabilistically, topics 13, 35, and 30 from the negative reviews are closest to topics 34, 37, and 33 from the positive reviews. Conversely, topics 49 and 23 from the negative reviews are the farthest from all positive review topics. Notably, topics 49 and 23 primarily contain words related to controls or input devices. This suggests that certain gameplay-related frustrations, such as unresponsive controls, have no corresponding themes in positive reviews. This highlights design issues that may alienate players even when other aspects of the game are appreciated.

4. Discussion

This section is split into two subsections. Section 4.1 presents the general discussion, while Section 4.2 details the limitations of the study.

4.1. General Discussion

Before addressing the topics themselves, it is worth noting that the results regarding review length align with those presented in the study of Lin et al. [31]. In their analysis of 28,159 reviews across several game genres, the median word count for the entire dataset was 30, with positive and negative reviews having medians of 19 and 40 words, respectively. In the case of English Soulslike reviews, the median was 7 words for positive reviews and 21 words for negative reviews. These results indicate that Soulslike players tend to write shorter reviews overall, although the standard deviation in word count is high for both review types. Nevertheless, the data indicate that review type significantly affects review length. This trend may be linked to the emotional and cognitive demands of the genre: positive experiences in Soulslike games often arise from feelings of mastery and immersion, which can be conveyed succinctly, whereas players who struggle with difficulty, controls, or bugs may be more inclined to describe their frustration in detail.

While Yu et al. conducted a topic modeling analysis of approximately 130,000 reviews focusing exclusively on Dark Souls: Remastered (DS1) and Dark Souls III (DS3) [55], the present study extends this scope by examining 418,483 English-language reviews across 21 different Soulslike titles available on Steam. This larger dataset enables a more comprehensive exploration of genre-wide patterns rather than game-specific phenomena. Yu et al.’s results identified 13 common topics, with high satisfaction related to items, boss fights, and combat themes, and persistent dissatisfaction regarding device issues, mainly controller functionality. In contrast, the current study not only confirms the salience of aesthetics, combat mechanics, and technical performance as central discussion points but also provides a more complex view of topic diversity across positive and negative reviews. Specifically, while Yu et al. reported positive rate statistics per topic, the present analysis highlights how topic clustering, intertopic distances, and the heterogeneity of negative feedback vary across the genre. This difference in scope and analytical depth allows the current work to capture both shared and divergent elements of player experience in Soulslike games, offering results that extend beyond the constraints of a single franchise and informing design considerations for a wider range of titles within the genre. Also, another study found significant differences between positive and negative feedback in terms of both emotion distribution and word usage [2]. The results of the present study corroborate these patterns, but they also expand them with additional temporal and contextual dimensions, showing, for example, how sentiment shifts over the course of the review text and how the agreement rate is influenced by review polarity and length. The results of the present study also expand on the conclusion that difficulty-driven frustration is a major emotional driver [1]. In this study, the multi-game analysis demonstrates that the intensity and valence of these emotions can differ significantly depending on the specific game, its sub-genre traits, and the community’s shared culture around it.

Regarding the topics shown in the present study, the two RQs were answered. RQ1 asked what themes are discussed in positive and negative reviews. The results showed that positive reviews commonly refer to visual design, world-building, and satisfaction derived from overcoming difficulty. Negative reviews focused more on gameplay mechanics, bugs, and user frustrations. RQ2 asked whether there are significant differences in topics between review types. The analysis revealed a clear separation in topic diversity and clustering: positive reviews were topically consistent, while negative ones were more varied. This suggests that player enjoyment in the genre is built around a shared understanding of its core appeal, while dissatisfaction arises from a broader range of issues.

The results show a difference between the diversity of topics in positive and negative reviews. Positive reviews are more similar to each other in terms of the topics covered. This suggests that players who enjoy these games similarly appreciate specific aspects. In contrast, negative reviews are less similar to each other, containing a wider range of topics. This indicates that the dissatisfactions of players differ between various aspects. Naturally, these diverse topics provide valuable information on specific areas, which can contribute to negative experiences for players. Topics can be grouped into five main categories. Group 1 is related to the Soulslike aesthetics, core elements, design, and graphics. Positive reviews highlight the visual and immersive elements of these games, indicating that players appreciate the atmosphere and the central elements of Soulslike games. Group 2 contains topics that are related to gameplay elements. While both types of reviews have topics related to these elements, negative ones put more focus on them. Players tend to highlight gameplay elements such as dodging, attacking, parrying, and other combat mechanics.

However, caution is needed when interpreting negative comments on gameplay mechanics in Soulslike games. Many of these systems are deliberately difficult and are intended to reward persistence and skill development. As such, negative reviews may sometimes reflect a player’s misalignment with the design philosophy, rather than a true mechanical flaw. Previous research shows that frustration is often part of the motivational loop in these games [5,7], and some critiques may stem from the genre’s uncompromising difficulty rather than poor design. This design intent is well understood by many developers working in the genre. Therefore, while comments about bugs or control responsiveness should be taken seriously, critical feedback regarding “hard” mechanics must be contextualized. Some developers may choose to preserve these elements because they form the core identity of Soulslike gameplay. This means that players who have had negative experiences pay closer attention to the core gameplay mechanics and may be more critical of the responsiveness of combat systems in these types of games. Group 3 has topics associated with feelings expressed in the reviews. Naturally, negative reviews tend to exhibit more negative emotions, indicating a sense of frustration or disappointment among the players who have had negative experiences. This also means that the player experience regarding game mechanics can shape the contents of reviews. Negative reviews also express regret over purchasing and spending money on the games. Group 4 contains topics related to bugs and issues. Negative reviews frequently mention technical problems and bugs encountered during gameplay, suggesting that the stability and performance of the games can affect player experience and, therefore, the reviews. Furthermore, negative sentiments toward the PC port and optimization suggest that players may have encountered specific problems related to the PC version of these games. Group 5 includes other topics that do not fall directly into the previous categories, although their numbers are small.

Although not directly extracted as a distinct topic, some negative reviews also referenced issues related to accessibility, such as lack of remappable controls, difficulty settings, or support for alternative input devices. These comments highlight the growing tension between preserving the challenge-based identity of the genre and opening it up to a more diverse player base. Future studies could benefit from more focused analysis of how accessibility concerns are voiced by players and how developers might address them without compromising core design goals.

Overall, the emergence of aesthetics- and atmosphere-related topics in positive reviews aligns with the Soulslike genre’s emphasis on environmental storytelling, intricate world-building, and immersive audiovisual design, all of which contribute significantly to player engagement. Conversely, the prominence of bug- and issue-related topics in negative reviews likely reflects the heightened expectations for technical stability in a genre already known for its high cognitive and emotional demands. The considerable presence of gameplay mechanics in negative reviews suggests that the steep learning curve and punishing difficulty, while core to the genre’s identity, can also be a source of frustration for certain players. This pattern supports the interpretation that both emotional and cognitive workload in Soulslike games strongly influence how players articulate their experiences in reviews.

The results have several implications for game developers and the research community. Firstly, they contribute to a deeper understanding of player experience with the Soulslike genre. Based on the results, developers can prioritize aspects of these games to create more engaging gaming experiences. For example, visual and atmospheric consistency appears to be a key strength in driving positive sentiment; maintaining high production quality in these areas may be a worthwhile investment. Similarly, negative reviews can help identify gameplay mechanics and technical issues, allowing developers to enhance and refine these identified elements. However, not all negative feedback should be interpreted as design failure, especially in a genre that intentionally frustrates. Genre conventions must be taken into account when deciding whether to modify mechanics or support systems. Furthermore, the contrasting patterns in the topics in both types of reviews can offer opportunities for further research. Researchers can study the relationship between specific elements of game design, player experience, and satisfaction. In return, they can formulate new hypotheses related to game design with the goal of improving player experience in the future. These results strengthen the fact that the Soulslike genre elicits a polarized form of player engagement, which is shaped by shared values of challenge and perseverance. The structure and diversity of review content across titles support the idea that positive feedback converges around genre-defining qualities, while negative sentiment is more fragmented and often personal. These distinctions make review analysis particularly useful for tracking how genre expectations are formed, challenged, and negotiated across time.

4.2. Limitations of the Study

Regarding limitations, this study has three primary ones. The first limitation is that only 21 Soulslike games were analyzed, as the reviews were originally scraped in 2021. Since then, more Soulslike titles have entered the market, including Elden Ring, the successor to the Dark Souls series developed by the same studio. As a globally successful and genre-defining title, Elden Ring represents a significant evolution in Soulslike design, blending open-world exploration with traditional combat systems. It attracted a larger audience, introduced greater mobility, and offered more flexible progression. Although Elden Ring was excluded due to its release date, this study provides a foundation for direct comparison. A follow-up investigation could determine whether the themes identified here persist or shift in light of more accessible, modern entries in the genre. In particular, it would be valuable to explore whether such hybridization has influenced player perceptions of difficulty, narrative, and mechanical depth.

The second limitation is that only English reviews were examined. English reviews were selected based on language tags assigned by Steam. However, some reviews tagged as English contained foreign words. Although these were relatively few, future research should filter them out to improve linguistic consistency.

The third limitation is that only reviews from the Steam digital distribution platform were analyzed. As a result, the dataset exclusively reflects the PC versions of these games. Including reviews from other digital game distribution platforms could yield more representative and comprehensive results.

5. Conclusions

According to the results, several key conclusions can be drawn. First, the number of words in a review is significantly affected by its type: positive reviews are shorter than negative ones. This may be because players who write negative reviews often feel the need to explain or justify their dissatisfaction, while those leaving positive reviews typically express approval without elaboration.

Second, the topics identified in the reviews can be grouped into five categories: Soulslike aesthetics and visual elements, gameplay mechanics, emotional responses, bugs and issues, and miscellaneous topics. Positive reviews most often highlight the games’ aesthetics, design, and overall experience, whereas negative reviews tend to focus on gameplay mechanics and bugs. This suggests that satisfaction is frequently rooted in environmental immersion and artistic coherence, while dissatisfaction commonly arises from functional or mechanical issues. Topic diversity analysis further showed that negative reviews cover a wider range of concerns, whereas positive reviews concentrate on a smaller set of shared strengths. In other words, players who enjoy the games value similar qualities, while those who dislike them cite varied and sometimes unrelated issues. This polarized pattern reinforces the Soulslike genre’s reputation as both highly rewarding and deeply divisive.

Third, these results have implications for game development. To create a similar game, developers should preserve the aesthetic and gameplay elements that consistently generate positive experiences. At the same time, addressing recurring complaints such as bugs and technical issues can help improve overall reception. Importantly, not all negative feedback should be interpreted as a design failure; some frustrations stem from intentionally challenging mechanics that define the genre. Developers must therefore balance preserving genre-defining difficulty with making accessibility adjustments that do not compromise the core experience.

Finally, this study provides a foundation for future research. The topics and patterns identified here can be used to assess new Soulslike games, including Elden Ring, which has received widespread critical acclaim and attention from both players and critics. Because the reviews analyzed in this study were written prior to Elden Ring’s release, these results offer a valuable baseline for comparison. A follow-up study could examine how Elden Ring reflects, modifies, or expands upon the themes identified here, thereby offering deeper insights into the evolution of the genre in response to larger audiences and hybridized mechanics (e.g., open-world exploration).

Overall, this paper not only reveals how players articulate their experiences with Soulslike games but also demonstrates the value of topic modeling in mapping the emotional, mechanical, and cultural dimensions of genre design and reception.