Sustainable Knowledge Contribution in Open Innovation Platforms: An Absorptive Capacity Perspective on Network Effects

Abstract

In open innovation platforms, users learn external knowledge through network interaction, and their position in the interactive network has an impact on the user’s sustainable knowledge contribution. Due to the gap in knowledge level, users’ absorption and utilization efficiency of external knowledge is not consistent. We studied the differences in user sustainable knowledge contribution behavior from the perspective of knowledge absorption. We crawled the data of a typical open innovation platform, used LDA to identify the level of user knowledge diversity and social network technology to analyze the user’s network location, and used the negative binomial regression model for empirical analysis. Our results show that knowledge diversity positively affects user knowledge contribution, and network breadth and network depth have positive and negative effects on user knowledge contribution behavior, respectively. In addition, the level of user knowledge diversity moderates the influence of network location on knowledge contribution. In summary, this research not only provides a comprehensive perspective on our current understanding of the contribution behavior of open innovation platforms, but also provides an in-depth understanding of how open innovation platforms can be properly designed to promote continuous contributions.

1. Introduction

With the rapid development of the information economy, the role of product users has changed, from a passive product receiver to an active value co-creator [1]. To take advantage of these user innovations, more and more companies have established an open innovative platform (OIP) to collect user ideas [2]. Open innovation was proposed by Chesbrough [3], and the central idea is to crowdsource knowledge to the public [4]; that is, user groups participate in creative tasks and collaborate to generate innovative ideas [5]. The open innovation platform is a platform sponsored by a company that uses the Internet as a medium, and is mainly aimed at providing an interactive space for companies and end users [6,7]. Users contribute knowledge to the company through the OIP. Companies acquire a large amount of external knowledge, develop knowledge and skills at a lower cost, and improve company innovation performance [8]. There are also some problems in the operation of OIP, and the number of ideas contributed by users is insufficient. Some OIPs have even been closed, such as IdeaStorm of Dell and My Starbucks of Starbuck. Therefore, improving user vitality and promote user idea contribution is a joint research problem of academics and enterprises.

The learning and creation of knowledge occur at the individual level, so effective management and rational use of individual knowledge are essential to the sustainable development of companies [9]. In OIP, user interaction is the main way for users to obtain external knowledge [10,11], and the network location of the user is very important for sustainable knowledge contribution [12]. From the perspective of social network analysis, the existing research believes that in a better position in the network, more and better knowledge of the platform can be obtained. However, different scholars have different judgments on the network location. In the interaction characteristics from the perspective of social networks, scholars believe that users in better network locations can obtain more external knowledge and thus provide more contributions; however, different scholars have different definitions of better network locations. For example, Peng et al. believe that extensive connections will result in more knowledge, thereby promoting user sustainable knowledge contributions [13], whereas Garriga et al. believe that integrating knowledge of extensive network connections will lead to coordination difficulties and reduce users’ motivation to contribute knowledge [14]. Freeman et al. believe that deep embedding of the network helps users to more quickly grasp all aspects of the platform’s knowledge [15], whereas Tang believes that if the user’s network is embedded too deeply, users will have less desire to explore new knowledge [16]. The possible reason for this is that these studies implicitly assume that knowledge acquired by users through network locations can be automatically absorbed, that is, knowledge acquisition is equivalent to knowledge absorption. Previous literature on OIP has noted the importance of network location for knowledge contribution from the perspective of social networks, but little attention has been paid to the importance of user knowledge absorption.

According to the knowledge absorption theory, internal knowledge constitutes the basis of individual knowledge absorption capacity and reflects the ability to absorb external knowledge [16]. The effective sharing and integration of internal knowledge is the key to knowledge absorption [17]. Therefore, for users with different knowledge levels, the influence of network location on sustainable knowledge contribution may be different, because users with different knowledge levels have different efficiencies in absorbing and using external knowledge. Examining only the impact of network location on users’ sustainable knowledge contribution behavior, while ignoring whether users’ knowledge levels are able to absorb external knowledge acquired through the network location, may yield an inaccurate impact on the perception of users’ sustainable knowledge contribution. Therefore, it is particularly important to study user sustainable knowledge contribution from the perspective of knowledge absorption. From the perspective of knowledge absorption theory, we explore the role of knowledge absorptive capacity of OIP users in the influence of network location on knowledge contribution. In our model, the user’s network location provides access to external knowledge beyond their knowledge boundaries, and the level of internal knowledge reflects the user’s ability to absorb external knowledge.

This article first reviews the previous work on OIP sustainable knowledge contribution, and discusses the knowledge absorption theory. Based on this, it proposes hypotheses and an integrated model. Then, this article reports the research background, methodology, analysis strategies, and empirical results, and conducts further robustness analysis. Finally, this article discusses the findings and impact of the research, and presents the limitations of the research. From a theoretical point of view, this study analyzes the influence of network location on users’ sustainable knowledge contribution, and explains the moderating effect of knowledge diversity on the influence from the perspective of knowledge absorptive capacity, which broadens the application field of knowledge absorption theory. From a practical perspective, this study applies machine learning text analysis and social network analysis to effectively identify and mine texts, expand valuable data sources for OIP users’ sustainable contribution behavior research, and relieve the cognitive load of manual work in enterprises.

2. Background

2.1. OIP Users’ Sustainable Knowledge Contribution

By providing precise requirements, users provide companies with innovative ideas and improve product quality [18]. Companies adopt high-value user innovative ideas and incorporate new product development to produce products that better meet user needs [2]. Users participating in platform-related knowledge exchange activities and contributing knowledge in an OIP can not only effectively achieve personal knowledge management goals, but also enhance the value of the platform, and user sustainable knowledge contribution in the OIP means that the user publishes ideas in the OIP [19]. User sustainable knowledge contribution is considered a key factor in the sustainable development of the platform [20].

Given the importance of users’ sustainable knowledge contribution, scholars have understood the influencing factors of this contribution from different angles. Previous research mainly explored the influencing factors of users’ sustainable knowledge contribution from three aspects: personal characteristics, interaction characteristics, and platform characteristics. (1) Personal characteristics are mainly based on the internal characteristics of users. For example, Nonaka [21] and Lai [22] believe that the level of user knowledge is extremely important for the contribution of user knowledge. Guan [23] introduced the importance of disclosing self-display information, and Khansa [24] also paid attention to aspects such as user active time. (2) Interaction characteristics refer to the characteristics formed in the process of external interactions between users and other people, such as comments, likes, and concerns, including interaction direction (positive and negative) [25], scale (number of interactions) [26], and network location ( social network characteristics such as degree centrality and intermediary centrality) [27]. (3) Little research exists on platform characteristics. For example, Hukal [27] studied the influence of platform support signals and opportunity signals on the contribution of platform users. This article only explores the sustainable knowledge contribution behavior of users on a single platform, and does not consider the impact of different platforms.

2.2. Knowledge Absorption Theory

Knowledge is an important source of a company’s core competitiveness [28]. From the perspective of a knowledge economy, the competition of companies has gradually transformed into a competition for knowledge [29]. Open innovation emphasizes the acquisition of external knowledge resources. In the application of external knowledge, it is particularly necessary to identify important knowledge and use it effectively [30]. Cohen proposed the absorptive capacity theory in 1990 to explain the dynamic changes in the stock of corporate knowledge [31]. Cohen believes that the key to a company’s knowledge stock is not the quantity at a certain point in time, but its ability to identify, transform, utilize, and develop from the environment, which is called “knowledge absorption capacity” [31]. Absorptive capacity is the synthesis of multiple abilities including knowledge acquisition, knowledge digestion, knowledge conversion, and knowledge application. The ultimate goal is to produce new knowledge [32].

With the continuous development of knowledge absorption theory, the level of analysis has also expanded from the organizational [33,34] to the individual [16,35]. Knowledge absorptive capacity at the organizational level represents the ability of an enterprise to identify, digest, and utilize new external knowledge in practice. Extended to the individual level, individual knowledge adsorptive capacity represents the individual’s ability to respond to knowledge, the process of fully combining external knowledge with internal knowledge, and transforming it into innovative output [35]. Knowledge absorptive capacity is an individual’s ability to respond to knowledge. Existing studies have explored the knowledge absorptive capacity in the field of open innovation, but the research objects are still companies and organizations [36], and relevant research on the individual level of OIP users has not yet appeared.

From the literature review, we found the following research gaps in the existing research. First, in the research on the influence of network location on knowledge contribution, different scholars have opposite views. Empirical research on network effects is mixed, making it difficult to advance theoretical understanding. Second, the research on knowledge absorptive capacity in the OIP field is only applied at the organizational level, ignoring the knowledge absorption of individual-level users. Knowledge absorptive capacity can be used as a theoretical perspective to explain this issue; however, in the existing research on OIP, the knowledge absorptive theory has not been applied to the user level.

We challenge this traditional view and try to fill this research gap in terms of two aspects. First, we consider both the positive and negative sides of network connections. We try to find theoretical mechanisms to explain the mixed network effects of user contributions. Second, we apply knowledge absorption theory to the individual level to explore the knowledge absorption capacity of OIP users, complementing the application of knowledge absorption theory. We assume that user knowledge contribution depends on knowledge absorptive capacity, and although network connectivity provides access to external knowledge, user knowledge contribution requires the moderating effect of user absorptive capacity. In an OIP, user innovation requires a series of knowledge related to product usage and service characteristics. Users establish interactive connections through the network to capture, learn, and absorb the knowledge of other users, and use this external knowledge to further contribute new knowledge. This study introduces the influence of users’ knowledge absorptive capacity on their sustainable knowledge contributions in open innovation platforms.

3. Research Model and Hypothesis Development

This section describes how the research model was developed (Figure 1). First, we explore the impact of network location, including network breadth and network depth, on user knowledge contribution, and propose hypotheses H1a and H1b. Second, we think that the level of knowledge diversity determines the user’s knowledge absorptive capacity, study the impact of knowledge diversity on the user’s knowledge contribution, and propose hypothesis H2. Finally, we explore the moderating effect of knowledge diversity in the influence of network location on user knowledge contribution, that is, the influence of network location on user knowledge contribution under the knowledge absorption effect, and propose hypotheses H3a and H3b. The hypotheses thus specify an important moderator—knowledge diversity—as guided by knowledge absorption theory, which has been largely understudied in the IS literature.

3.1. The Influence of Network Location on Users’ Sustainable Knowledge Contribution

Social learning theory believes that social members will have learning behaviors under the influence of others, and this influence may be direct interaction or indirect observation [37]. In an OIP, users can increase their skill reserves by interacting with other users, thereby enhancing creativity and promoting sustainable knowledge contributions. The interaction in an OIP creates a network relationship to obtain external knowledge. In the paper, network breadth refers to the degree of connection between users and other users, and network depth refers to the level of embeddedness in the platform’s interactive network.

Network breadth refers to the range of knowledge gained by users through network connections. Through the interactive network, users can directly connect with other users in the network and obtain external knowledge of other users. The social capital theory shows that the collection of all knowledge resources in a person’s relationship network can strongly affect the degree of interpersonal knowledge sharing [38]. Knowledge is very important for innovation, but the cost of acquisition is very high. The interaction between members of the OIP provides a cost-effective way to obtain a wider range of knowledge sources. The more social interactions users have, the greater the intensity, frequency, and breadth of information exchange [39]. The more connections a user has on the network, the more innovative resources and external information they can access during the innovation process. The more innovative knowledge exchanges with other users, the higher the innovation performance may be [40]. Extensive network connections can provide users with new insights, reduce cognitive effort, and increase the rate of creation of new knowledge [13].

Hypothesis 1a (H1a):

Network breadth positively affects user sustainable knowledge contribution.

Network depth refers to the contact distance with other users in the interactive network, that is, the degree of embeddedness in the network. In other words, network depth means the distance from a user to other users connected directly or indirectly in the network. When deeply embedded in the network, users can quickly access relevant information. However, an excessively embedded network will cause users to access redundant information, which limits their ability to effectively explore new knowledge in the network [41]. Proximity to other users’ network locations may cause similar or redundant information loops, confining users to their perceptions. This kind of cognitive lock-in may inhibit users’ motivation to explore new knowledge from external networks and hinder their motivation to continue to innovate, resulting in a decline in sustainable knowledge contribution. When the user is deeply embedded in the network, the user’s motivation to explore new ideas and create new knowledge from the network will be weakened. As old knowledge becomes obsolete, users further lose their motivation to continue to innovate, which hinders their sustainable knowledge contribution [16].

Hypothesis 1b (H1b):

Network depth negatively affects user sustainable knowledge contribution.

3.2. The Influence of Knowledge Diversity on User’s Sustainable Knowledge Contribution

Knowledge diversity refers to the abundance of individual knowledge, experience, and skills, and is a measure of the user’s internal knowledge level [42]. The collection of knowledge elements owned by each individual and the relationship between these collections constitute a personal knowledge base. Innovation is the process of reorganizing the knowledge elements in the knowledge base [43]. In an OIP, users have different professional levels and experience, and each person’s knowledge inventory is also different, and there is a gap in their innovative ability. Users with diverse knowledge are more able to promote knowledge transfer and sharing [44]. According to Nonaka’s research, diverse knowledge will stimulate users’ innovative thinking and produce more practical ideas [21].

Users with different knowledge levels have different motivations to contribute to their knowledge. Generally speaking, users with low knowledge levels only contribute knowledge to obtain platform rewards, whereas users with high levels not only gain platform revenue but also gain new knowledge [44]. According to the theory of planned behavior, in an OIP, users’ perception and control of knowledge creation are determined by their knowledge level. If users do not have enough knowledge, even if they have the willingness to create new knowledge, they will not undergo sustainable knowledge contribution behavior [45]. In other words, the willingness to innovate alone is not enough, and the knowledge and ability to support the generation of innovation is also required [46]. In an OIP, users need a series of knowledge related to products and services to propose ideas.

Hypothesis 2 (H2):

The diversity of user knowledge positively affects user sustainable knowledge contributions.

3.3. Difference Analysis of Knowledge Absorption Effect to User Sustainable Knowledge Contribution

Absorptive capacity is the ability to recognize, digest, transform, and develop and utilize knowledge. Cognitive and behavioral science research shows that absorbing knowledge is the process of using the knowledge through the evaluation of external knowledge, establishing connections with pre-concepts, and associating existing knowledge after possessing internal knowledge [16]. The process of interactive digestion of external knowledge and internal knowledge is the process of knowledge absorption. The knowledge absorption effect expresses the degree of utilization of knowledge after the interaction between internal knowledge and external knowledge [16]. For OIP users, based on existing internal knowledge, they can use external knowledge by establishing network connections with other users.

According to cognitive load theory, each user’s attention is limited. When internal knowledge is highly diversified, the value of acquiring external knowledge may be more limited, because a wealth of internal knowledge can provide enough new perspectives. The high network breadth will increase the cost for users to integrate knowledge from different sources. This diverse external knowledge may collide with the existing internal knowledge, resulting in difficult coordination [47]. When users accept a wide range of external knowledge, the direction of internal knowledge may be the opposite. Users need to spend more time and energy to coordinate this knowledge, which will significantly reduce the speed of the sustainable knowledge contribution. For users with a high level of knowledge, an extensive network may also increase the complexity of the integration of internal and external knowledge, which reduces the level of user knowledge absorption and leads to a decrease in user sustainable knowledge contribution [48]. For users with low knowledge levels, extensive network connections can help users obtain more external knowledge, and the problem of insufficient internal knowledge can be alleviated through extensive external knowledge.

Hypothesis 3a (H3a):

Knowledge diversity negatively moderates the influence of network breadth on sustainable knowledge contributions.

Deeply embedded networks can obtain overall network information more easily and quickly, and achieve better performance with “less information transmission, shorter time, and lower cost” [15]. The extensive knowledge base helps users evaluate development trends from different perspectives. When deeply embedded in the network, users can quickly develop the best knowledge about technological trends and related expertise [15,16], and obtain the best development direction. Although excessive network embedding may cause knowledge redundancy for users, for users with high knowledge levels with diverse knowledge, their broad knowledge and multi-angle thinking can alleviate this cognitive lock-in. The deeply embedded network helps users better and more quickly understand the overall status of the OIP network and grasp the latest trends [16]. The smaller the knowledge distance, the fewer intermediate steps the knowledge receiver takes to absorb knowledge, and the efficiency of knowledge absorption increases. For users with high knowledge levels, by quickly acquiring external knowledge, understanding the most promising development direction, and gathering their diversified knowledge, they can create more new knowledge more quickly. For users with low levels of knowledge, based on insufficient internal knowledge, deep embedding in the network makes it worse. Redundant external knowledge can only limit the development of their thinking and is not conducive to their sustainable knowledge contribution.

Hypothesis 3b (H3b):

Knowledge diversity positively moderates the influence of network depth on sustainable knowledge contributions.

4. Research Methodology

4.1. Research Context and Data Collection

Trailblazer Community (https://trailblazers.salesforce.com accessed on 1 January 2020), an OIP of SalesForce, was selected as the research sample. SalesForce is one of the world’s largest manufacturers of customer relationship management software (CRM). To facilitate users to proposes ideas, SalesForce created an OIP called Trailblazer Community for registered users to add, propose, correct, and comment on various ideas.

This study chose Trailblazer Community as the context to test the hypothesis, because: (1) Trailblazer Community is one of the most typical OIPs, and the platform is mature and has a large number of registered users; (2) Trailblazer Community is a comprehensive open innovation platform. In addition to the IdeasExchange section, it also adds an Answers section (users can ask each other questions), providing more useful user-generated content for the research; (3) the website has typical OIP components, including user profile interfaces, an idea content interface, and representative social network content.

To ensure the activeness of users, users who published ideas and answers within one year (June 2019–June 2020) were selected as the research objects in June 2020 (T1). The user behavior data was crawled through Python, including personal information, creative comment status, Q&A status, and other data to obtain explanatory variables. A total of 1469 members who met this criterion were included in the analysis. After obtaining the explanatory variables for 12 months, the users’ participation data were re-crawled for the explained variables in June 2021 (T2). In this way, the endogenous problem was remedied. In addition, information from users who had published ideas within two years (June 2018 to June 2020) was also crawled as a robustness test.

4.2. Measures

4.2.1. Dependent Variable

The purpose of this study was to analyze the sustainable knowledge contribution of users in an OIP. Users provide new knowledge by publishing ideas in an OIP. Therefore, the dependent variable should be measured by the ideas of users in the OIP. The number of sustainable knowledge contributions is represented by the number of ideas published by a user.

4.2.2. Independent Variables

(1)

Knowledge Diversity

The level of user knowledge diversity is a relatively difficult indicator to measure. In previous studies, the measurement of knowledge diversity mostly used a questionnaire method, self-test, or expert review [16]. However, a huge amount of knowledge is generated every day in an OIP, and the methods of questionnaire or expert review will consume a large amount of manpower, material resources, and financial resources. Text mining and machine learning are used to measure the level of user knowledge. As information topics are not pre-categorized, this research draws on the knowledge classification approach proposed by Hwang [41], using natural language processing technology, i.e., Latent Dirichlet Allocation (LDA), to solve this problem. LDA is a three-level Bayesian probability graph model, and its composition structure contains three granularities of the document, topic, and term [49]. The relationship between document, topic, and term is shown in Figure 2.

In the figure, K is the number of topics, M is the total number of documents, and N is the total number of words in the m-th document. β is the Dirichlet prior parameter of the polynomial distribution of the term under each topic, and α is the Dirichlet prior parameter of the polynomial distribution of the topic under each document. Z_mn is the topic of the n-th word in the m-th document, and W_mn is the n-th word in the m-th document. The two hidden variables θ_m and φ_k, respectively, represent the distribution of topics under the m-th document and the distribution of words under the k-th topic. The former θ_m is a k-dimensional (k is the total number of topics) vector, and φ_k is a v-dimensional vector (v is the total number of terms in the dictionary).

Based on these definitions and notations, LDA assumes that the generation process of each text document in the corpus is as follows:

Step 1: If there are 1–m documents and 1–k topics, then use a polynomial distribution (Dirichlet distribution) for each document (m) to randomly assign it to each topic (k). For example, the probability of document A being assigned to topic 1 is 25%, the probability of being assigned to topic 2 is 25%, and the probability of being assigned to topic 3 is 50%.

Step 2: If there are 1–n words, then each word belongs to a certain topic (k) with a certain probability. For example, the word “mean” belongs to the topic “statistics” with a probability of 0.25.

Step 3: For each word (n) in the document (m) and topic (k), calculate the proportion of the word in the document to the topic, and record it as the probability P(k|m) of the topic (k) in the document (m). Then calculate the proportion of the word (n) in the topic (k) from all documents containing the word, and record it as the probability P(n|k) that the word (n) belongs to the topic (k).

Step 4: Re-sampling, the benchmark probability here is P(k|m) × P(n|k).

Step 5: Repeat this process; after many iterations, the algorithm will eventually converge. Therefore, the document is assigned to a topic based on the proportion of each word assigned to the topic in the document.

In the selection of the number of topics, we use the perplexity evaluation index to determine the optimal number of topics in the document. Perplexity is commonly used to measure the advantages and disadvantages of a probability distribution or probability model prediction sample, which can be used to adjust the number of topics [50]. The calculation formula of perplexity is shown in (1).

$$\mathrm{perplexity}\left(D\right)=\exp \frac{{{\displaystyle \sum }}_{d=1}^M\log P\left(W_d\right)}{{{\displaystyle \sum }}_{d=1}^M{N}_d}$$

(1)

where D represents the set of all words in the document; M represents the number of documents; W_d represents the words in document _d; N_d represents the number of d words in each document; P(W_d) represents the probability of occurrence of the words in the document. The perplexity value generally shows a decreasing law with the increase in the number of potential topics. The smaller the perplexity value, the stronger the generation ability of the topic model. Therefore, we determine the optimal number of topics in the model to be 75 based on perplexity.

Then, the number of main topics per user is calculated based on the results of the LDA. The distribution of response documents for each user under 75 topics was obtained by LDA. According to the method proposed by Hwang [41], topics with a topic distribution of more than 0.013 were recorded as main topics (total distribution probability of 1 for 75 topics, or 0.013 for each topic distribution if the topics were evenly distributed). The level of user knowledge diversity is measured by the number of main topics.

(2)

Network Breadth and Network Depth

A social network analysis approach was used to study variables related to the user’s network location by creating a network of user comments. All ideas posted by users over the course of a year were captured, along with the commenters of those ideas. Pajek was then used to build a network of comment relationships between users and commenters, resulting in an undirected graph having 4635 nodes and 9263 edges. The network relationship graph is shown in Figure 3.

Network breadth. Network breadth is measured using a degree centrality with reference to Tang [16]. From the perspective of social networks, the higher the degree of centrality of a node, the more important the connections in this network. Therefore, the degree center performance is a good measure of network breadth. The standardized degree centrality measurement formula proposed by Scott [50] is used to obtain the proportion of node i connected to other nodes. The calculation of degree centrality is shown as Formula (2):

$$\mathrm{degree} \mathrm{centrality}=\frac{{{\displaystyle \sum }}_{j=1}^N{x}_{ij}}{N-1}\left(i\ne j\right)$$

(2)

where N is the total number of users in the network and j represents other users connected to i.

Network depth. Network depth is measured using a closeness centrality with reference to Tang [16]. From the perspective of social networks, closeness centrality reflects the closeness of a node to other nodes. The higher the closeness centrality of a node, the closer it is to other nodes and the easier it is to reach other nodes, and it can access the information in the network faster and more accurately [51]. Therefore, the closeness centrality is a good measure of the extent to which users are embedded in the network, that is, the depth of the network. The closeness centrality calculation method is proposed by Freeman in 1979 [15], as shown in Formula (3):

$$\mathrm{closeness} \mathrm{centrality}=\frac{N-1}{{{\displaystyle \sum }}_{j=1}^{N}d\left(g_{i,j}\right)}$$

(3)

where N is the total number of users in the network and $d\left(g_{i,j}\right)$ is the geodetic (shortest) path between users i and j, which is determined by the minimum number of links between i and j.

As the values obtained in the previous step were too small, to ensure the visibility of the results, the above values were normalized using the Spark MinMaxScaler method.

4.2.3. Control Variables

To exclude the control variables of individuals, the control variables were chosen from two perspectives—interaction characteristics and personal characteristics—according to the Salesforce website design and existing literature. Differences in interaction characteristics refer to the different behaviors shown by users in the process of socializing with other users on the platform, and are expressed by the numbers of followers and people being followed [52]. Differences in individual characteristics refer to different behaviors in the innovation process due to the user’s ability, quality, or participation, which are expressed by self-expression [23] (how many words in their introduction) and tenure [53].

Table 1. Descriptive statistics of variables.

Construct	Measure Item	Description	Mean	Std. Dev.	Min	Max
DV (measured at T2)
Sustainable knowledge contribution	Idea	The number of ideas of user in T2	3.671	11.355	1	193
IVs (measured at T1)
Knowledge diversity	Knowledge	The number of main topics included in the user answer text	7.026	4.468	0	28
Network Breadth	NetworkBreadth	Degree centrality of user idea-comment network	0.018	0.050	0	1
Network Depth	NetworkDepth	Closeness centrality of user idea-comment network	0.019	0.045	0	1
Control Variables (measured at T1)
Following	Following	The number of following users	3.838	25.330	0	535
Followers	Followers	The number of followers of users	8.200	78.117	0	1703
Self-expression	Introduction	The number of self-displayed information by users	4.592	3.898	0	10
Control Variables (measured at T2)
Tenure	Tenure	The number of days from the user’s first idea to T2	326.792	75.381	30	365

shows the descriptive statistics of all variables. The standard deviation of each variable represents the degree of dispersion of the variable. The larger the standard deviation, the more discrete the variable distribution; the smaller the standard deviation, the more concentrated the variable distribution. Min and Max represent the minimum and maximum values of the variable in the sample. For example, knowledge diversity is measured by counting the number of main topics in the user’s answer text. The Min value of 0 means that the text does not contain any topic, and the Max value of 28 means that there are 28 main topics in the text.

Table 2. Correlation matrix and VIF value.

Variable	V0	V1	V2	V3	V4	V5	V6	V7	VIF
V0 Ideas	1.000
V1 Knowledge	0.380	1.000							4.110
V2 NetworkBreadth	0.176	0.148	1.000						4.072
V3 NetworkDepth	0.128	0.048	0.591	1.000					1.852
V4 Following	0.135	0.107	0.222	0.101	1.000				1.856
V5 Followers	0.170	0.068	0.484	0.463	0.501	1.000			1.063
V6 Introduction	−0.250	−0.105	−0.142	−0.119	−0.050	−0.058	1.000		1.060
V7 Tenure	−0.102	−0.062	−0.037	0.006	−0.036	−0.068	0.179	1.000	1.058

shows the correlation matrix of all variables. The values in the correlation matrix are all less than 0.7, indicating that the correlation between variables can be ignored. To avoid multicollinearity, we performed the VIF test, and the results are shown in the last column of Table 2. All values are less than 5, indicating that there is no obvious multicollinearity problem.

4.3. Model Specification and Estimation

The dependent variable is a count variable, which is characterized by non-negative integers and over-dispersion [54] (i.e., Table 1 reveals that the mean of the dependent variable differs greatly from its variance). The negative binomial regression (henceforth NB) is employed to test the hypotheses. In addition, there is an interaction term in the model, so a hierarchical regression model was used to examine the main and moderating effects. The regression model is as follows:

$$\begin{array}{l}Ideas={\alpha }_0+{\alpha }_{1}Knowledge+{\alpha }_{2}NetworkBreath+{\alpha }_{3}NetworkDepth+{\alpha }_{4}Knowledge\times NetworkBreath\\ +{\alpha }_{5}Knowledge\times Depth+{\alpha }_{6}Following+{\alpha }_{7}Follower+{\alpha }_{8}Introduction+{\alpha }_{9}Tenure\\ +{\epsilon }_i\end{array}$$

(4)

where ${\alpha }_0$ is the intercept term, ${\epsilon }_i$ is the error term, and ${\alpha }_i$ (i = 1, 2, …, 9) is the coefficient of each explanatory variable (including the interaction term) and the control variable.

5. Results

5.1. Estimation Results

The results are presented in

Table 3. Regression results.

Variables	Model 1	Model 2	Model 3
Knowledge	0.151 ***	0.130 ***	0.122 ***
	(0.005)	(0.005)	(0.006)
NetworkBreadth	4.723 ***	3.958 ***	6.655 ***
	(1.002)	(0.896)	(2.020)
NetworkDepth	−2.404 **	−2.721 **	−9.039 ***
	(1.209)	(1.075)	(2.617)
Knowledge*NetworkBreadth			−0.334 *
			(0.201)
Knowledge*NetworkDepth			0.789 ***
			(0.289)
Following		0.000	0.000
		(0.001)	(0.001)
Follower		0.001 *	0.001 *
		(0.000)	(0.000)
Introduction		−0.059 ***	−0.058 ***
		(0.007)	(0.007)
Tenure		−0.002 ***	−0.002 ***
		(0.000)	(0.000)
Constant	−0.165 ***	0.815 ***	0.887 ***
	(0.050)	(0.113)	(0.115)
Observations	1468	1468	1468
Log likelihood	−3089.159	−3004.569	−3000.497
Pseudo R-square	0.131	0.155	0.156

Notes: Robust standard errors in parentheses. * p < 0.1. ** p < 0.05. *** p < 0.01.

. Model 1 only considered the main effects for NB; Model 2 added control variables based on Model 1; and Model 3 tested the moderating effects by adding interaction items. In addition, Model 3 has the highest goodness of fit, so Model 3 with interaction terms is the most suitable model.

The data analysis and measurement results are shown in Table 3. Negative binomial models were estimated by maximum likelihood estimation rather than least squares estimation, so R-square is meaningless. We also report Pseudo R-square in the last row. Larger Pseudo R-square values indicate stronger model explanatory power. Table 3 indicates that Model 3 with all variables has the strongest explanatory power.

(1)

The influence of user knowledge diversity on sustainable knowledge contribution is positive and significant (Model 3, β = 0.122, p < 0.01). That is, the more knowledgeable a user, the more ideas he will publish, and H1 is supported.

(2)

There is a positive correlation between user network breadth and sustainable knowledge contribution (Model 3, β = 6.655, p < 0.01). This shows that the more interactive network connections the user has, the more ideas can be generated, and the network breadth has a great influence on the contribution of knowledge. Hypothesis H2a is verified.

(3)

There is a negative correlation between user network depth and sustainable knowledge contribution (Model 3, β = −9.039, p < 0.01), indicating that the greater the user’s embeddedness in the interactive network, the less the user’s creative ideas, and hypothesis H2b is verified.

(4)

Knowledge diversity negatively moderates the positive correlation between network breadth and sustainable knowledge contribution (Model 3, β = −0.334, p < 0.1). This shows that, for users with a high level of knowledge, establishing more network connections will not further promote the publication of more ideas, but will inhibit users’ enthusiasm for innovation. This supports Hypothesis H3a.

(5)

Knowledge diversity positively moderates the relationship between network depth and sustainable knowledge contribution (Model 3, β = 0.789, p < 0.01), which shows that for users with high knowledge levels, deeper network embedding can promote create more ideas, which validates Hypothesis H3b.

5.2. Robustness Checks

To verify the robustness of the study results, the data collection and measurement models were further tested for robustness.

First, this was undertaken for the range of data selection. As mentioned earlier, users who had published ideas within one year were previously selected in order to ensure that OIP users were active. In the robustness test, users who had published their ideas within two years were selected as the study population for the same NB regression analysis. The results are shown in

Table 4. The results of robustness checks.

Variables	Model 4	Model 5	Model 6
Knowledge	0.121 ***	0.140 ***	0.143 ***
	(0.004)	(0.003)	(0.002)
NetworkBreadth	47.920 ***	8.711 ***	30.59 ***
	(11.350)	(0.690)	(3.453)
NetworkDepth	−3.342 ***	−7.344 ***	−6.521 ***
	(1.005)	(1.011)	(0.736)
Knowledge*NetworkBreadth	−0.270 **	−0.591 ***	−0.637 ***
	(0.123)	(0.072)	(0.047)
Knowledge*NetworkDepth	0.611 ***	0.531 ***	0.612 ***
	(0.107)	(0.102)	(0.046)
Following	−0.127 ***	0.000	−0.0621 ***
	(0.032)	(0.001)	(0.010)
Follower	−0.074 **	0.001 ***	−0.036 ***
	(0.033)	(0.000)	(0.010)
Introduction	−0.006	−0.068 ***	−0.027 ***
	(0.007)	(0.004)	(0.004)
Tenure	0.000	−0.003 ***	−0.000 ***
	(0.000)	(0.000)	(0.000)
Constant	0.134 *	1.140 ***	0.177 ***
	(0.074)	(0.052)	(0.039)
Observations	2908	1468	2908
Log likelihood	−5861.466	−4379.681	−8625.744
Pseudo R-square	0.127	0.432	0.302

Notes: Robust standard errors in parentheses. * p < 0.1. ** p < 0.05. *** p < 0.01.

, Model 4. It can be seen from Table 4 that the significance of the regression results of the two-year data and the one-year data is the same, and all hypotheses were verified.

Second, the measurement model was tested. The Poisson regression model was used in place of the NB model for regression analysis. The results of users who published ideas within one year are shown in Table 4, Model 5, and the results of users who published ideas within two years are shown in Table 4, Model 6. As can be obtained from the table, the results are the same as before, indicating that the proposed model is robust.

6. Discussion

This paper introduced the knowledge absorption theory, explored the knowledge absorption effect of users in the open innovation community, and analyzed how the user’s internal knowledge diversity level, external network interaction capabilities, and the interaction between internal knowledge and external networks affect sustainable knowledge contribution.

6.1. Theoretical Discussion

This article made some theoretical contributions.

First, this study contributes to the literature on open innovation platforms. Specifically, our study provides a complementary introduction to previous studies examining individual characteristics of user idea contribution in OIPs. Previous research found that users who occupy dominant network locations sustainably contribute knowledge. In the process of exploring OIP knowledge contribution, our research introduces a new perspective of absorptive capacity, links the internal knowledge level and the external network location, and breaks down the barriers between users’ internal knowledge and external interaction.

Second, this study extends existing knowledge absorption theories. Existing knowledge adoption theories, such as SECI or the I-space knowledge management model, emphasize the distinction, integration, and transformation of “tacit knowledge” (knowledge that is not easy to express and difficult to normalize) and “explicit knowledge” (knowledge that can be encoded and expressed in words or numbers). From the perspective of social network analysis, based on the external network and internal knowledge level, we explored the user’s ability to absorb external network knowledge under different knowledge levels. Building on knowledge assimilation theory, we highlight the differential impact of internal knowledge levels on external network locations, extending the application of knowledge assimilation theory at the individual level.

Finally, this study explores the positive and negative results of network connectivity and embeddedness, and the results provide a warning for researchers who only examine the impact of network connectivity from the perspective of social networks without considering internal capabilities. The research proves that it is necessary to balance internal and external knowledge to create more new knowledge. This echoes the equilibrium theory of absorptive capacity and provides important empirical evidence to support the equilibrium viewpoints proposed in previous theories.

6.2. Managerial Implications

There are also some managerial implications in this article. Furthermore, we offer recommendations to companies using OIPs.

First, for companies that have invested heavily in OIPs, our work provides management guidance for managers to deploy people to take advantage of their network connections. A user with a balanced external network connection is more likely to contribute more knowledge. Wider network connectivity, or reduced network embedding, can encourage users to contribute more knowledge. By facilitating discussion among users in the community, such as incentivizing user comment behavior, the breadth of user network connections can be increased. In addition, by reducing exposure to popular items, the redundancy of user knowledge can be weakened, thereby reducing the embeddedness of the community network and the depth of network connections.

Second, the results of this study can provide a reference for companies to collect user-generated data. Companies such as Best Buy, Unilever, General Electric, and PlayStation have all hosted OIPs, but some OIPs cannot attract users to contribute sustainably, and even close down, such as those of Dell and Starbucks. Therefore, it is very important to attract users to make sustainable high-quality contributions. Our results show that, for companies that host an OIP, by collecting data on the micro-activity of users’ knowledge contributions, the company can identify those individuals who have the potential to innovate more ideas. By mobilizing individuals with knowledge in a wide range of fields, companies may be able to increase their knowledge contribution through an OIP.

Finally, the same guidance applies to the management of idea contributors, but with caution, since OIP project managers have less control over these users. For platforms and companies, creating more network connections is not necessarily a better incentive for users to innovate; this, differentiating between users with different levels of knowledge is required. For users with extensive knowledge, too many connections are not conducive to sustainable knowledge contribution, and gaining deeper connections allows the user to access external knowledge more quickly and easily. For users with limited knowledge, more connections can alleviate the deficiency caused by the lack of knowledge; establishing deeper connections can discourage them from creating new knowledge.

6.3. Limitations and Future Research

Nonetheless, this study has some limitations. First, due to the platform settings, the comments cannot be displayed at specific points in time, and can only be expressed in the form of “months ago” and “years ago”, making it impossible to use the panel data for more accurate research. Second, the measurement of variables is somewhat subjective and needs further improvement and refinement in the future. Finally, this study only empirically explores OIPs for CRM software, and subsequent ongoing research can attempt more types of OIP.

Our research provides avenues for further research. First, due to the limitations of the empirical research, we only used website objective data to explore knowledge absorption in user knowledge contributions. Future research can use a combination of qualitative and quantitative methods to examine the role of these theoretical mechanisms in the continued contribution of OIP users. Second, future research can mine dynamic features through panel data to more accurately explore user sustainable knowledge contributions. Third, future research can choose different types of OIP and compare the differences in users’ sustainable contribution behaviors in different industries and types of OIP.

7. Conclusions

This research attempts to develop and empirically test theoretical models that integrate different levels of user diversity, network connectivity, and embeddedness to improve the current understanding of users’ sustainable knowledge contributions. The study proposes a comprehensive model based on knowledge uptake theory and validates it against large-scale data collected from a typical OIP. The findings suggest that users with different levels of knowledge diversity have different abilities to absorb knowledge through web locations, and that understanding these synergies is necessary.