ANALYSIS OF SOCIAL MEDIA DATA TO OPTIMIZE THE MARKETING STRATEGY OF GSOM HIGHER EDUCATION PROGRAMS
|
ЗАЯВЛЕНИЕ О САМОСТОЯТЕЛЬНОМ ХАРАКТЕРЕ ВЫПОЛНЕНИЯ ВЫПУСКНОЙ КВАЛИФИКАЦИОННОЙ РАБОТЫ 2
STATEMENT ABOUT THE INDEPENDENT CHARACTER OF
THE MASTER THESIS 3
Introduction 3
Chapter 1. Use of social media for higher education promotion and problem statement 5
1.1. History of approaches to higher education promotion 5
1.2. Social media platforms' popularity 7
1.3. Problem description 8
1.4. Machine Learning approach in social media analytics 11
1.5. Research goals and expected results 14
Chapter 2. Exploratory data analysis and methodology for working with text data 16
2.1. Methodology of text processing 16
2.2. Data description and transformations 23
2.3. Dashboards for tracking main estimators 30
Chapter 3. Building Machine Learning models for VK data processing 34
3.1. Dataset preparation for model building 34
3.2. Modelling the dependencies between likes and post characteristics 36
3.3. Cluster analysis 46
Conclusion 52
Reference list 54
STATEMENT ABOUT THE INDEPENDENT CHARACTER OF
THE MASTER THESIS 3
Introduction 3
Chapter 1. Use of social media for higher education promotion and problem statement 5
1.1. History of approaches to higher education promotion 5
1.2. Social media platforms' popularity 7
1.3. Problem description 8
1.4. Machine Learning approach in social media analytics 11
1.5. Research goals and expected results 14
Chapter 2. Exploratory data analysis and methodology for working with text data 16
2.1. Methodology of text processing 16
2.2. Data description and transformations 23
2.3. Dashboards for tracking main estimators 30
Chapter 3. Building Machine Learning models for VK data processing 34
3.1. Dataset preparation for model building 34
3.2. Modelling the dependencies between likes and post characteristics 36
3.3. Cluster analysis 46
Conclusion 52
Reference list 54
In recent years, social media has become an increasingly important tool for promoting higher education programs. As more and more students turn to social media to research colleges and universities, it has become essential for institutions to have a solid social media presence.
Social media has changed the higher education marketing landscape in an ambivalent way. On the one hand, it allows institutions to reach a wider audience and connect with prospective students in new ways and rather efficiently thanks to technologies of targeting advertisement. Universities can leverage social media data to create more focused content and engage with the right audience (Chaudhry, 2023). On the other hand, there are difficulties with getting the target audience to notice the content and engage with it. More than 30% of social media professionals struggle with it (Calus, 2023), and the educational sphere is undoubtedly not an exception. Colleges and universities are constantly creating content, but the question is how to actually reach the target audience. Usually, higher education establishments use social media to connect with prospective students by sharing information about academic programs, campus events, and student life and addressing concerns and questions. However, institutions sometimes struggle with creating engaging content that resonates with their target audience. If they manage to do so, it will help people feel like a part of the university culture and encourage them to share the posts further, therefore expanding the reach/invoking word-of-mouth marketing (Rogers).
Overall, social media can be a powerful tool for promoting higher education programs if institutions use it correctly. They should create informative, entertaining, visually appealing content that aligns with the institution's brand and builds a solid social media presence. Institutions can show their advantages to their target audience and attract more prospective students. Moreover, it is crucial for institutions to use social media strategically and to tailor their approach to the needs and interests of their potential students.
Our work consists of the following parts: Chapter 1 covers the history and trends of education marketing and describes the GSOM marketing strategy, particularly in the Russian social media network - VK. Moreover, the first chapter discusses the latest ML methods for evaluating social media marketing performance. Chapter 2 describes the methodology we chose for our research and VK API data used for data analysis. Also, the second chapter covers the dashboard design and its managerial use. Chapter 3 discusses the models for posts and followers’ analysis and its main results. In Conclusion, we provide data-driven recommendations to the marketing department and discuss areas for further research.
Generally, the work was distributed inside of the group equally. For Introduction and the first chapter, each participant wrote an equal volume of text, which was then discussed and revised together. The exploratory data analysis and data preparation was done by all the authors jointly. For the practical part the work was divided in the following way: A. D. Lisitsyna was responsible for the dashboard and all the work connected to it, M.O. Ryleeva built classification models and made k-means clusterization, and D.A. Doroshkova built regression models and DBScan based clusterization. Conclusions from the analysis were drawn and written together.
Social media has changed the higher education marketing landscape in an ambivalent way. On the one hand, it allows institutions to reach a wider audience and connect with prospective students in new ways and rather efficiently thanks to technologies of targeting advertisement. Universities can leverage social media data to create more focused content and engage with the right audience (Chaudhry, 2023). On the other hand, there are difficulties with getting the target audience to notice the content and engage with it. More than 30% of social media professionals struggle with it (Calus, 2023), and the educational sphere is undoubtedly not an exception. Colleges and universities are constantly creating content, but the question is how to actually reach the target audience. Usually, higher education establishments use social media to connect with prospective students by sharing information about academic programs, campus events, and student life and addressing concerns and questions. However, institutions sometimes struggle with creating engaging content that resonates with their target audience. If they manage to do so, it will help people feel like a part of the university culture and encourage them to share the posts further, therefore expanding the reach/invoking word-of-mouth marketing (Rogers).
Overall, social media can be a powerful tool for promoting higher education programs if institutions use it correctly. They should create informative, entertaining, visually appealing content that aligns with the institution's brand and builds a solid social media presence. Institutions can show their advantages to their target audience and attract more prospective students. Moreover, it is crucial for institutions to use social media strategically and to tailor their approach to the needs and interests of their potential students.
Our work consists of the following parts: Chapter 1 covers the history and trends of education marketing and describes the GSOM marketing strategy, particularly in the Russian social media network - VK. Moreover, the first chapter discusses the latest ML methods for evaluating social media marketing performance. Chapter 2 describes the methodology we chose for our research and VK API data used for data analysis. Also, the second chapter covers the dashboard design and its managerial use. Chapter 3 discusses the models for posts and followers’ analysis and its main results. In Conclusion, we provide data-driven recommendations to the marketing department and discuss areas for further research.
Generally, the work was distributed inside of the group equally. For Introduction and the first chapter, each participant wrote an equal volume of text, which was then discussed and revised together. The exploratory data analysis and data preparation was done by all the authors jointly. For the practical part the work was divided in the following way: A. D. Lisitsyna was responsible for the dashboard and all the work connected to it, M.O. Ryleeva built classification models and made k-means clusterization, and D.A. Doroshkova built regression models and DBScan based clusterization. Conclusions from the analysis were drawn and written together.
During our research, many experiments have been conducted using different models and sets of parameters. The third chapter describes all the models and their results in detail. This section will focus on the conclusions drawn from the models' results.
First of all, a substantial number of followers are not active and irrelevant to the analysis, they are not the target audience. For example, the pages that provide services for writing papers or fulfilling different tests for students cannot be considered a target audience of GSOM. It is important to remember that fact when measuring engagement and maybe use an alternative metric instead of standard ER that takes into account the total number of followers. For example, the number of likes divided by the number of views can be a good substitute. This finding highlights the need to promote higher education programs and not let things take their course, as we indicated in our work before. That is why the launch of target advertisement in VK that the marketing department already plans is necessary. Our findings made during this research can help set up the target advertisement campaign more efficiently. The marketing department can use the information derived from cluster analysis to tailor targeting ads content to the interests of the specific cluster of users. In particular, we would advise focusing on extracurricular activities provided by the university in correspondence with the interests of the targeted cluster. Also, we would recommend adding the university ratings in the ads as it is important for enrollees, according to the results of our research.
What is more, we would recommend the marketing department try to "bring the groups back to life" not only by acquiring but also by keeping the followers who are real people interested in enrolling in GSOM. For this purpose, there is a need to make content that is interesting for the target audience. We found out that while the presence of video content matters, it is not the primary factor when we look at the followers' reactions. It is the text of the post that is more important. So, we would recommend focusing on the content, in particular, to tailor it to the audience's preferences that can be tracked on the dashboard. For example, as the dashboard and cluster analysis show, the most widespread interests among our followers are travel and sport. Hence, the marketing department should write about student exchange, languages and sports opportunities that GSOM provides to attract the attention of the target audience. Moreover, during our research, we discovered some "trigger" words that significantly influence engagement. Based on this information, we would recommend writing about GSOM events, clubs, and news, as well as business topics and partner companies. Also, the best posting time is in the morning and evening hours, at this time, the posts are more likely to be seen and reacted to. All the recommendations described above should help the marketing departments achieve their goals and attract prospective enrollees to VK groups.
However, it is worthy of note that our research has some limitations. First of all, the type of data used is time series, hence, there is a risk of structural changes in the data. For instance, a significant shift in the number of likes (from tens to hundreds) could make our models less relevant. Secondly, the amount of data collected is not that big (three thousand posts from all three groups), which could involve a risk of model overfitting and limit our ability to derive more detailed conclusions from the data. That is why we recommend continuing research after applying all the recommendations we provided. Further steps could be gathering more data - at least from two complete cycles of admission campaigns (one before applying the recommendations and one - after) to test the new content efficiency and check the target advertisement results. Moreover, it would be interesting to look at the data of competitors' groups' activity to gain some new insights and find out how GSOM can stand out among other universities. Also, some A/B tests can be conducted when testing the new way of creating the group content. Another area for improvement is looking at the differences between engaged and not engaged followers. Furthermore, qualitative interviews could be conducted with several group participants - this would allow for generating hypotheses for the ML task.
First of all, a substantial number of followers are not active and irrelevant to the analysis, they are not the target audience. For example, the pages that provide services for writing papers or fulfilling different tests for students cannot be considered a target audience of GSOM. It is important to remember that fact when measuring engagement and maybe use an alternative metric instead of standard ER that takes into account the total number of followers. For example, the number of likes divided by the number of views can be a good substitute. This finding highlights the need to promote higher education programs and not let things take their course, as we indicated in our work before. That is why the launch of target advertisement in VK that the marketing department already plans is necessary. Our findings made during this research can help set up the target advertisement campaign more efficiently. The marketing department can use the information derived from cluster analysis to tailor targeting ads content to the interests of the specific cluster of users. In particular, we would advise focusing on extracurricular activities provided by the university in correspondence with the interests of the targeted cluster. Also, we would recommend adding the university ratings in the ads as it is important for enrollees, according to the results of our research.
What is more, we would recommend the marketing department try to "bring the groups back to life" not only by acquiring but also by keeping the followers who are real people interested in enrolling in GSOM. For this purpose, there is a need to make content that is interesting for the target audience. We found out that while the presence of video content matters, it is not the primary factor when we look at the followers' reactions. It is the text of the post that is more important. So, we would recommend focusing on the content, in particular, to tailor it to the audience's preferences that can be tracked on the dashboard. For example, as the dashboard and cluster analysis show, the most widespread interests among our followers are travel and sport. Hence, the marketing department should write about student exchange, languages and sports opportunities that GSOM provides to attract the attention of the target audience. Moreover, during our research, we discovered some "trigger" words that significantly influence engagement. Based on this information, we would recommend writing about GSOM events, clubs, and news, as well as business topics and partner companies. Also, the best posting time is in the morning and evening hours, at this time, the posts are more likely to be seen and reacted to. All the recommendations described above should help the marketing departments achieve their goals and attract prospective enrollees to VK groups.
However, it is worthy of note that our research has some limitations. First of all, the type of data used is time series, hence, there is a risk of structural changes in the data. For instance, a significant shift in the number of likes (from tens to hundreds) could make our models less relevant. Secondly, the amount of data collected is not that big (three thousand posts from all three groups), which could involve a risk of model overfitting and limit our ability to derive more detailed conclusions from the data. That is why we recommend continuing research after applying all the recommendations we provided. Further steps could be gathering more data - at least from two complete cycles of admission campaigns (one before applying the recommendations and one - after) to test the new content efficiency and check the target advertisement results. Moreover, it would be interesting to look at the data of competitors' groups' activity to gain some new insights and find out how GSOM can stand out among other universities. Also, some A/B tests can be conducted when testing the new way of creating the group content. Another area for improvement is looking at the differences between engaged and not engaged followers. Furthermore, qualitative interviews could be conducted with several group participants - this would allow for generating hypotheses for the ML task.
