New Business Models, Values, Explorative mapping review, topic modeling, LDA, LSA,
Abstract
This extended abstract reports the research design and preliminary findings of our exploratory mapping review on the research presented at the New Business Model Conference (NBMC). We describe the rationale for an exploratory mapping review, our supervised and unsupervised approach (text mining, topic modeling, Latent Dirichlet allocation (LDA), Latent Semantic Analysis (LSA)) to analyzing the conference proceedings, the preliminary findings, and our next steps. This study aims to examine the used definition, conceptualization, and contextualization of “ new business models” and “ value creation” within the presented research to understand recent research development better, identify emerging trends, assess the conference’s predictive value based previous research, and draw lessons for the field's future. Our analyses will help to show how words evolve and relate to each other, which helps us look back, reflect, look forward, and predict, which can serve as a stepping stone to conducting a more rigorous systematic literature review and will result in a better understanding of the definitions and conceptualization regarding “ new business models” and “ multiple value creation” as well as their linkages In addition, we hope to gain insights from and align the interdisciplinary contributions to new business model research and multiple value creation, identify areas of overlap, and determine the need for theoretical and practical synthesis.
Introduction
What do “ New Business Models” and “ Multiple Value Creation” mean? What words do we use to describe them, and what is their relationship? What can we learn from the research topics and quality presented at NBM, which gaps exist, and how did this research help practitioners? These questions and our desire to better understand the influence of multiple value-creation forces on (new) business models kick-started this exploratory literature review and topic modeling.
Multiple value creation involves simultaneously achieving multiple economic, social, and environmental objectives through organizational activities that engage relevant stakeholders. With growing awareness of balancing financial performance with social and environmental responsibility, this concept has gained significant attention in recent years. As a result, multiple value creation (MVC) has become a central concept of interest for researchers and practitioners focussing on sustainability, circular economy, social entrepreneurship, environmental awareness, and innovation (Breuer, Ludeke-Freund, & Bessant, 2022; Freudenreich, Lüdeke-Freund, & Schaltegger, 2020; Jonker & Faber, 2019; Lumpkin & Bacq, 2019; Saebi, Foss, & Linder, 2019; Tapaninaho & Kujala, 2019).
This paper aims to conduct a mapping review of the concept of multiple value creation by utilizing R for data analysis. While the diverse application of MVC has contributed significantly to our understanding of this concept, it has also led to fragmentation. To address this, an exploratory mapping review effectively synthesizes and integrates existing literature (Asmussen & Møller, 2019; Munn et al., 2018). Our sample includes conference proceedings from the "New Business Models" conference held over the past six years, which we believe provides a valuable source of information for exploring the concept of MVC.
Munn et al. (2018, p. 2) define mapping reviews as a tool to determine the scope of a specific topic's body of literature to identify and map the available evidence. The authors continue that this review type provides insight into the available evidence in a field, clarifies key concepts and definitions, examines how research is conducted, identifies the main characteristics or factors related to a concept, and might be a precursor for a systematic review. A scoping review seeks to "explore and define conceptual and logistic boundaries around a particular topic to inform a future predetermined systematic review or primary research." (Sutton, Clowes, Preston, & Booth, 2019, p. 211). We use (un)supervised machine learning (topic modeling, LDA, and LSA) to accomplish this. Asmussen and Møller (2019) indicate that applying LDA topic modeling to exploratory literature reviews is a promising but underutilized approach.
Latent Semantic Analysis (LSA) is a computational approach to understanding the contextual meaning of words. It entails a computational technique that combines automated content analysis and information retrieval to provide researchers with a more objective approach to analyzing textual data (Evangelopoulos, Zhang, & Prybutok, 2012). By analyzing a vast collection of text (a corpus) using statistical methods, LSA can extract and represent the shared meaning of words based on the contexts in which they commonly occur (Landauer, Foltz, & Laham, 1998). Whereas Wagire, Rathore, and Jain (2020) note that LSA has been extensively employed in knowledge extraction studies, as demonstrated by its incorporation in several literature review papers, such as those authored by Sidorova et al. (2008), Sidorova and Isik (2010), Marksberry et al. (2011), Kundu et al. (2015), and Lin et al. (2017). We apply this technique to learn from the conference proceedings and better understand “new business models” and “multiple value creation.”
Hence, we have multiple objectives for conducting this research:
- Better understand the meaning of the words we use to describe our concepts (conceptualization and definition)
- To better understand the relationships between the words used to research business models and value creation
- to provide a comprehensive overview of the existing literature on MVC based on the conference proceedings and how it relates to the broader field;
- assess the type of research and the quality of the research conducted;
- to identify gaps in the existing literature and make recommendations for future research;
- and lastly, to detect patterns that could be extrapolated to the future.
Background
The concept of multiple value creation is multifaceted and closely tied to the emergence of new business models. However, the nature of this relationship remains unclear. To gain a deeper understanding of this phenomenon, it is imperative first to define multiple value creation and new business models. The conference proceedings of the New Business Models Conference are utilized to examine the development and interplay of these concepts.
In recent years, numerous review articles have tackled the definitional ambiguity surrounding "new business models" and "value creation.” These articles generally agree that the field is characterized by fragmentation, diversity, and pluralism regarding output, focus, and design (Kraus, Filser, Puumalainen, Kailer, & Thurner, 2020; Massa, Tucci, & Afuah, 2017). This was confirmed by Mina and Michelini (2022) in last year’s conference proceeding. They identified trends and developments in business model research, including increased attention to sustainability and a call for intellectual debate about the development of the field.
The research areas on business models continue to evolve, with disruptive events such as COVID, technological advancements like IT/AI, and changing customer or user preferences influencing the composition of business models and their value creation accents (Schwidtal et al., 2023; Sewpersadh, 2023; Taherdoost & Madanchian, 2023).
Method
Our study began with identifying relevant keywords to develop a comprehensive codebook. We employed text mining techniques, including topic modeling using Latent Dirichlet Allocation (LDA) and Latent Semantic Analysis (LSA), on converted PDF documents using the open-source software R. This approach enabled us to gain insight into the frequently used terms in each document and to assess the word frequency, as well as the correlation between the terms and associated concepts(Feinerer, 2013; Silge & Robinson, 2017). Furthermore, we utilized the concept function in Atlas to identify relationships between text phrases in the documents. To facilitate our analysis, we used the frequently found words for auto-coding in Atlas.ti, which will help to triangulate the findings (Benchimol, Kazinnik, & Saadon, 2022; Wild, 2007). We are conducting the LDA and LSA analyses, of which we will describe the procedure in detail below.
LDA
Asmussen and Møller (2019) indicate that applying LDA topic modeling to exploratory literature reviews is a promising but underutilized approach. They defined the method as a probabilistic, unsupervised technique for topic modeling that identifies topics within a set of papers. This method defines a topic as a probability distribution over a fixed vocabulary. LDA analyzes the words in each document and determines the joint probability distribution between the observed (words in the document) and the unobserved (the underlying structure of topics). To achieve this, LDA employs a 'Bag of Words' approach, where the meaning and semantics of sentences are not considered, and only the frequency of words is evaluated. We followed the approach of Amussen, which consists of the following steps:
Figure 1: Process Overview Amissen p.6
1. Data collection: We retrieved the conference proceedings from the website. As the PDF files had several issues in the source code, we saved every page as a pgn, merged these, and applied an OCR to recognize the text.
2. Pre-processing and Term Reduction. We conducted the standard pre-processing steps, such as: converting to lower case, removing punctiation, numbers, English stopwords, and white space. Additionally we developed a custom list of stopwords to remove unimportant words, for example: "nbm", "proceedings", "july". Please see Appendix X for the complete list of stopwords used.
a. Remove tikes with less than two characters (do)
b. We ran an analysis with and without stemming
3. Topic Modelling – We are currently in this stage
1. Post Processing. As Asmussen and Møller (2019) wrote, “The post-processing steps aim to identify and label research topics and topics relevant for use in a literature review. An outcome of the LDA model is a list of topic probabilities for each paper. The list is used to assign a paper to a topic by sorting the list by the highest probability for each paper for each topic. By assigning the papers to the topics with the highest probability, all of the topics contain similar papers (p. 6-7). Labeling the topics we are currently working on and presenting during the conference follows.
a. Validation of the results (to be executed)
We used the descriptions of Asmussen and Møller (2019) on Github as well as that of Welbers, Van Atteveldt, and Benoit (2017) (also on Github) and the tutorial on myBinder to conduct the analysis.
LSA
LSA is a technique for processing language that facilitates identifying relationships between a collection of documents and the terms contained therein. Specifically, LSA detects similarities and differences in the usage of words and phrases within each document and across all documents in a given period, enabling the extraction of conceptual meanings from a body of text, such as a set of conference proceedings consisting of many short papers on various topics. As described by Boukus and Rosenberg (2006) in Mazis (2017), the output of LSA represents the document set's major associative patterns or underlying textual themes, achieved by associating words and phrases based on their frequency of occurrence within and across the document (Mazis & Tsekrekos, 2017, p. 5)
To conduct the LSA, we followed the approach of (Wagire, Rathore, & Jain, 2020), visualized in Figure 2 underneath.
Figure 2: Process overview Wagire et al. (2020)
Steps 1 and 2 are the same as described in the previous section on LDA
1. Data collection: We retrieved the conference proceedings from the website. As the PDF files had several issues in the source code, we saved every page as a pgn, merged these, and applied an OCR to recognize the text.
2. Pre-processing and Term Reduction. We conducted the standard pre-processing steps: converting to lowercase, removing punctuation, numbers, English stopwords, and white space. Additionally, we developed a custom list of stopwords to remove unimportant words, for example: "nbm", "proceedings", and "july". Please see Appendix X for the complete list of stopwords used.
a. Remove tikes with less than two characters (do)
b. We ran an analysis with and without stemming
The following steps need to be executed
3. TDM transfromation into TF-IDF
4. SVD
5. Factor extraction and factor rotation
6. Factor interpretation and labeling.
Preliminary Results
We are currently working on the LDA, which is an interactive process that asks for a continuous back-and-forth with the outcomes and the data. For instance, the parameters must be assessed and the stopwords altered based on the outcomes. See the tables underneath to provide an illustration the LDA outcomes based on 12 topics (see the Appendix for 25 topics) - which both need to reassed and adapted (remove words that do not add anything or have been lost in convertion as well as running several analyses with adapted parameters). We also used the “ldatuning::FindTopicsNumber” function to assess the number op topics needed. Furthermore, we have not yet completed the steps listed in the previous section. We cannot stress enough that these figures are merely illustrative.
Figure 3 Example Topics LDA
This can be graphically represented in the following manner
Figure 4 Example of Distribution LDA Topics over document
De documents were uploaded from 2016-2022, indicating D1 = 2016, and so forth. This preliminary result shows the relationship of the topics with the documents. As indicated, the data needs to be processed again so this image is merely for illustrative purposes of the eventual functionality and use instead of truly informative contentwise.
Figure 5 Example Clusters LSA
Conclusion
We trust that our research provides an exciting opportunity to learn from the NBM research contributions and conceptualization of the terms used to describe (new) business models and multiple value creation (capute and delivery). Currently, we are making sense of the data and can only show descriptive outcomes (success as word frequency, concepts, word clouds and correlational findings on the frequently used words). During the conference, we will present our updated methodology and methods employed and our finished analyses.
References
Asmussen, C. B., & Møller, C. (2019). Smart literature review: a practical topic modelling approach to exploratory literature review. Journal of Big Data, 6(1), 1-18.
Benchimol, J., Kazinnik, S., & Saadon, Y. (2022). Text mining methodologies with R: An application to central bank texts. Machine Learning with Applications, 8, 100286.
Breuer, H., Ludeke-Freund, F., & Bessant, J. (2022). Editorial — Special Issue: Managing Values for Innovation. International Journal of Innovation Management, 26(05), 2201001. doi:10.1142/s1363919622010010
Freudenreich, B., Lüdeke-Freund, F., & Schaltegger, S. (2020). A stakeholder theory perspective on business models: Value creation for sustainability. Journal of Business Ethics, 166(1), 3-18.
Jonker, J., & Faber, N. (2019). Business models for multiple value creation: Exploring strategic changes in organisations enabling to address societal challenges. Sustainable business models: innovation, implementation and success, 151-179.
Kraus, S., Filser, M., Puumalainen, K., Kailer, N., & Thurner, S. (2020). Business Model Innovation: A Systematic Literature Review. International Journal of Innovation and Technology Management, 17(06), 2050043. doi:10.1142/s0219877020500431
Lumpkin, G. T., & Bacq, S. (2019). Civic Wealth Creation: A New View of Stakeholder Engagement and Societal Impact. Academy of Management Perspectives, 33(4), 383-404. doi:10.5465/amp.2017.0060
Massa, L., Tucci, C. L., & Afuah, A. (2017). A critical assessment of business model research. Academy of Management annals, 11(1), 73-104.
Mazis, P., & Tsekrekos, A. (2017). Latent semantic analysis of the FOMC statements. Review of Accounting and Finance, 16(2), 179-217.
Munn, Z., Peters, M. D., Stern, C., Tufanaru, C., McArthur, A., & Aromataris, E. (2018). Systematic review or scoping review? Guidance for authors when choosing between a systematic or scoping review approach. BMC medical research methodology, 18(1), 1-7.
Saebi, T., Foss, N. J., & Linder, S. (2019). Social entrepreneurship research: Past achievements and future promises. Journal of management, 45(1), 70-95.
Schwidtal, J. M., Piccini, P., Troncia, M., Chitchyan, R., Montakhabi, M., Francis, C., . . . Kiesling, L. (2023). Emerging business models in local energy markets: A systematic review of peer-to-peer, community self-consumption, and transactive energy models. Renewable and Sustainable Energy Reviews, 179, 113273. doi:https://doi.org/10.1016/j.rser.2023.113273
Sewpersadh, N. S. (2023). Disruptive business value models in the digital era. Journal of Innovation and Entrepreneurship, 12(1). doi:10.1186/s13731-022-00252-1
Silge, J., & Robinson, D. (2017). Text mining with R: A tidy approach: " O'Reilly Media, Inc.".
Sutton, A., Clowes, M., Preston, L., & Booth, A. (2019). Meeting the review family: exploring review types and associated information retrieval requirements. Health Information & Libraries Journal, 36(3), 202-222.
Taherdoost, H., & Madanchian, M. (2023). Blockchain-Based New Business Models: A Systematic Review. Electronics, 12(6), 1479.
Tapaninaho, R., & Kujala, J. (2019). Reviewing the stakeholder value creation literature: Towards a sustainability approach. In Social responsibility and sustainability (pp. 3-36): Springer.
Wagire, A. A., Rathore, A., & Jain, R. (2020). Analysis and synthesis of Industry 4.0 research landscape: Using latent semantic analysis approach. Journal of Manufacturing Technology Management, 31(1), 31-51.
Welbers, K., Van Atteveldt, W., & Benoit, K. (2017). Text Analysis in R. Communication Methods and Measures, 11(4), 245-265. doi:10.1080/19312458.2017.1387238
Wild, F. (2007). An LSA package for R. Paper presented at the Proceedings of the 1st International Conference on Latent Semantic Analysis in Technology Enhanced Learning (LSA-TEL'07).
Appendix I
Example LDA 25 topics
Appendix II
Outcome topicfinder
Appendix III
Topic list based on Binder analysis (top7termsPerTopic)
Creative Commons Attribution 4.0 International License (CC-BY 4.0)
