Financial Services Review, 33(2) 36 Exploring the Effect of Federal Student Loan Payment Resumption on Borrowers Through Sentiment and Textual Analysis Using X Jason N. Anderson,1 Donovan Sanchez,2 Juan E. Gallardo,3 Derek Lawson,4 and Congrong Ouyang5 Abstract Student loans have taken on an increasingly significant role in funding the higher education experience and payments toward student loan debt have become an important part of many borrowers’ overall financial plan. Using Brandwatch, this study analyzes X data to better understand student loan borrower sentiment during the resumption of federal student loan payments in October 2023. During the period studied, negative references to student loans on the platform overtook positive sentiment overwhelmingly (46% negative versus 1% positive). Topics and phrases labeled negative sentiment ranked higher in mentions than those with positive or neutral sentiment in their respective categories. The findings highlight the need for financial planners to provide appropriate mental health resources to help borrowers manage negative feelings surrounding the federal student loan payment restart. Creative Commons License This work is licensed under a Creative Commons Attribution-Noncommercial 4.0 License Recommended Citation Anderson, J., Sanchez, D., Gallardo, J., Lawson, D., & Ouyang, C. (2025). Exploring the effect of federal student loan payment resumption on borrowers through sentiment and textual analysis using X. Financial Services Review, 33(2), 36-54. Introduction Given the increasing costs of college education, many college attendees rely on student loans to complete their degrees. According to the National Center for Education Statistics (2023), the average total cost for a first-time, full-time undergraduate student living on campus at a public 4-year institution during the 2021-2022 1 Corresponding author (jasonanderson@ksu.edu). University of Kansas, Lawrence, Kansas, USA. 2 Kansas State University, Manhattan, Kansas, USA. 3 Kansas State University, Manhattan, Kansas, USA. 4 Kansas State University, Manhattan, Kansas, USA. 5 Kansas State University, Manhattan, Kansas, USA. Disclosure: We employed AI algorithms, specifically Brandwatch, to analyze the data collected for this study. The AI methods were selected based on their suitability for the research objectives and data characteristics. academic year was $26,000. For first-time, full- time undergraduate students attending private nonprofit 4-year institutions, the price tag was much higher at $55,800. From March 13, 2020, to September 1, 2023, the U.S. Department of Education paused payments and set interest rates to 0% on eligible federal student loans as part of the nation’s COVID-19 https://creativecommons.org/licenses/by-nc/4.0/ mailto:jasonanderson@ksu.edu https://creativecommons.org/licenses/by-nc/4.0/ Anderson et al. 37 emergency relief measures. During this same period, the Biden Administration proposed a plan to forgive up to $20,000 for qualifying borrowers. This plan was ultimately blocked by the U.S. Supreme Court in June 2023. After this decision, student loan interest was set to begin accruing in September 2023 with payment resumption in October 2023. The breakneck pace of student loan changes over the last four years has undoubtedly affected student loan borrowers, who must now address their student loan debt in addition to cumulative societal transitions brought on by the pandemic. This project seeks to understand student loan borrower sentiment during payment resumption through an analysis of their expressions on social media. An increased understanding of borrower sentiment will help financial professionals better appreciate the mental health ramifications of this financial transition on borrowers. The authors wish to make a point of clarification in light of company name changes that could create confusion for the reader. In July 2023, Elon Musk rebranded the well-known microblogging platform, Twitter, to X. Because this change is so recent, however, most previous studies will refer to the platform by its former name, Twitter. To avoid confusion, the authors have chosen to refer to the platform as X in this paper, even retroactively, unless Twitter is referenced directly in a quote or article title. The purpose of this research is to evaluate the emotional impact of recent student loan changes on borrowers. One way to gauge the emotional change within student loan borrowers is through the proxy of textual sentiment during times of major student loan changes. To proxy student loan borrowers, researchers can measure individuals who discuss student loans on X. As mentioned previously, from early 2020 to late 2023 student loan borrowers made no payments or accrued interest on eligible federal student loans, and the Biden Administration proposed a generous loan forgiveness program. After the denial of student loan forgiveness, interest accrual resumed in September 2023 and payments in October 2023. This paper seeks to address the following question: How did sentiment change on X during the resumption of federal student loan payments in October 2023? Based on the findings of Sinha et al. (2023) – which are covered later in the literature review - we hypothesize that: H1: Negative sentiment will outweigh positive sentiment during the study’s date range. H2: Topics and phrases with negative sentiment will rank higher in mentions than those with positive or neutral sentiment during the study’s date range. Literature Review Student Loans Because student loans are readily accessible to students, generally have lower interest rates than private loans, and offer fixed rates, it is no surprise that their usage has increased (Ebrahimian, 2023). Despite their popularity, borrowers have polarized views about student loans. As a means of funding their education, some borrowers feel that the cost is worth it, while others feel weighed down (Nuckols et al., 2020). Such feelings often prompt individuals to share their student debt experiences on social media platforms. Student loans have been shown to have numerous effects on borrowers. For example, Kim and Chatterjee (2019) identified a negative association between student loans and an individual’s life satisfaction and psychological well-being. Student loans have also been associated with psychological distress, financial anxiety, and other negative forms of health and psychological well-being (Archuleta et al., 2013; Kim & Chatterjee, 2019; Zhang et al., 2019). Student loans have also had negative outcomes to the extent that some individuals develop problems with mental health, smoking, and heavy drinking (Qian et al., 2021). Not all borrowers view student debt the same way. Borrowers in the repayment period of their loans reported higher psychological distress than those who were still enrolled in a university (Sato et al., 2020). In addition, and when compared to individuals without student loans, those repaying their loans had lower levels of financial satisfaction (Kim et al., 2021). Reflecting contradictory findings, Joseph and MacDonald Financial Services Review, 33(2) 38 (2021) concluded that student loans are negatively associated with financial satisfaction, while Robb et al. (2018) found no significant effect. The contradicting results from these studies highlight the complexity of the relationship between student loans and the emotional state of borrowers. In addition to the psychological effects associated with student loan debt, asset accumulation has also been highlighted in research. Mountain et al. (2020) found that student loans negatively affected Millennials’ homeownership rates. Also of concern, student debt has been associated with a reduced likelihood to save for retirement (Elliott et al., 2013). Beyond concerns related to retirement and housing, student loan borrowers are more likely to carry other types of debts, like car and credit card debt (Fry, 2012). Debt repayment reduces disposable income for borrowers, and the resulting lack of liquidity can make it more challenging to finance other purchases (Gicheva & Thompson, 2014). Furthermore, having student debt hinders borrowers’ ability to spend. In a recent paper, 18% of student loan holders reported difficulty buying daily necessities because of their existing debt (Hanson, 2023). Social Media as a Data Source In contrast to traditional research sources, social media data is “user-generated, naturalistic, and unstructured,” creating a fascinating opportunity for researchers to explore public opinion on a variety of topics (Sinha et al., 2023, p. 736). Social media as a tool to express opinions has become more popular as the number of users has increased with the availability of the internet (Ortiz-Ospina, 2019). While social media may be used to share various types of information, it has also been used to express sentiment on specific topics that affect a wide range of individuals. Considering its massive growth, companies and media organizations are interested in exploring X user data as a means of detecting user sentiment on various products and services (Kouloumpis et al., 2011). Likewise, significant interest has developed among researchers for purposes of better understanding the public’s opinion on important topics. What makes X (and social media in general) advantageous from a research standpoint is that it provides insight into the minds of users by the voluntary posting of their thoughts, expressions, as well as their interactions with other platform users in “a naturalistic setting” (De Choudhury et al., 2013, p. 128) and in real-time (Edo-Osagie et al., 2020). Chancellor et al. (2020) note that new computational methods of social media data analysis could make significant differences, such as using social media data to identify and provide interventions for risky behavior. The significant number of users also provides a benefit to researchers in that there are a large sample of tweets to look at. As Zimbra et al. (2018, p. 24) notes, “Many researchers and firms have recognized that valuable insights on issues related to business and society may be achieved by analyzing the opinions expressed in the abundance of tweets”. The Use of Social Media Data in Research Harvesting social media data is a non-traditional means of obtaining the opinions and sentiment of subjects. Traditional forms of gathering opinion and sentiment data include the use of surveys to try to determine how study participants feel about a particular issue. The internet and modern technologies have created new spaces for researchers to gather data (Edo-Osagie et al., 2020). In their research seeking to identify depression on X, Nadeem et al. (2016) found that social media data represents a potential solution to problems that can arise in self-reported depression questionnaires in that postings on social media often provide a window directly into the state of mind of the social media user. Researchers can use social media data to “automatically identify self-expressions” in the construction of a given data set (Coppersmith et al. 2014, p. 52). Limited research has been conducted to date using social media and X for analysis of student loans. One notable exception is a 2023 study by Sinha et al. looking at student loans and mental health expressions on Reddit and X. Sinha et al. (2023) used Scarcity Theory to explore Reddit and X data to improve understanding of the relationship between student loans and mental health. They found that social media posts about Anderson et al. 39 student loans were associated with negative sentiment and had a higher likelihood of containing expressions of sadness and fear. A large body of work used X and other social media data to study mental illness and mood disorders. In their review of studies attempting to predict mental illness, Guntuku et al. categorized studies into two camps: finding correlates to, or identifying, mental illness (2017, p. 43). Preoţiuc- Pietro et al.’s (2015) research focused on predicting “linguistic markers” of mental illness (p. 28) while De Choudhury et al. (2013) predicted major depression using X. Similarly, Chancellor et al. (2020) used social media data to identify mood and psychosocial disorders. Reece et al. (2017, p. 8) used X to predict depression and PTSD, noting that their “method identified these mental health conditions earlier and more accurately than the performance of trained health professionals, and was more precise than previous computational approaches”. Finally, Coppersmith et al. (2018) demonstrated how automatic procedures using social media data and Natural Language Processing (NLP) can detect suicide risk. Considering the unique way individuals use social media, researchers are naturally attracted to it as a means of examining a variety of behaviors, moods, opinions, and sentiment. Because X allows users to express themselves and interact with others, researchers can monitor sentiment and analyze data with relative ease. Social science explores the interaction of individuals in their environments with other human beings, and X documents these interactions in a way that can be informative to the social science researcher. In their research, Quantifying Mental Health Signals in Twitter, Coppersmith et al. noted that “social media is by nature social, which means that social patterns, a critical part of mental health and illness, may be readily observable in raw Twitter data.” (2014, p. 51). Coppersmith et al. (2014) successfully used X data to perform individual and population-level analyses to identify mental health disorder signals for depression, bipolar disorder, post-traumatic stress disorder, and seasonal affective disorder. In another study, Coppersmith et al. were able to “easily and automatically” identify X users with PTSD by “scanning for tweets expressing explicit diagnoses” instead of relying on “traditional PTSD diagnostic tools” (2014, p. 579). Tsugawa et al. (2015, p. 9) found that by analyzing X user history data that “depression can be recognized in users with an accuracy of approximately 69%”. While text is typically the object of analysis for researchers, Guntuku et al. looked at image postings and profile pictures on X for clues relating to user mental health and found that anxious users tended to post more photos, and both depressed as well as anxious users posted images “dominated by grayscale” (2019, p. 244). Edo-Osagie et al. (2020) undertook a scoping review of the use of X for public health research and found that studies most often used X data for surveillance, event detection, pharmacovigilance, forecasting, disease tracking, and geographic identification. Sentiment Analysis The vastness of social media calls for useful summarization tools to gain insights into the underlying data. Sentiment analysis can automate this otherwise cumbersome process to pull opinionated data from massive datasets (Giachanou et al., 2017). Sentiment analysis strives to improve the automatic recognition of sentiment within a text (Zimbra et al., 2018) and detects opinions based on features selected by researchers (Giachanou et al., 2017). While there are a variety of ways to seek to understand student loan sentiment among borrowers, sentiment analysis offers a unique opportunity. As noted previously, automating the processing of large chunks of data is valuable from an efficiency and effectiveness standpoint. Most beneficial, in the opinion of the authors, is the opportunity to understand borrower sentiment as expressed by them without prompting from a researcher. In a way, sentiment analysis allows researchers to observe human behavior in its natural habitat. X Sentiment Analysis X is a microblogging platform that allows users to publish their opinions on virtually any topic. Social science researchers can examine user sentiment with relative ease on a wide range of Financial Services Review, 33(2) 40 issues and across a large population considering X’s substantial user base and significant number of daily messages (Giachanou et al., 2017). X sentiment analysis may be viewed as a “specialized area within sentiment analysis” (Zimbra et al., 2018, p. 3) where X data is mined for opinionated text on a given topic (Giachanou et al., 2017), with sentiment detection occurring through the establishment of word embeddings (Carvalho et al., 2021). X sentiment analysis is not without its challenges. Some of these challenges include length limitation of messages (Giachanou et al., 2017), “novel language” resulting from length of message constraints (Zimbra et al., 2018, p. 24), as well as written errors and content that is constantly changing (Giachanou et al., 2017). Despite its challenges, X sentiment analysis remains an important tool for researchers. It offers direct insights from users, shedding light on public opinions on critical issues (Giachanou et al., 2017). Methodology On June 23, 2023, X removed academic API access and significantly lowered the caps allowed for data collection by academics. This development made it nearly impossible for academics to gather data in a cost-effective manner for use in open-sourced statistical software or readily available data science models. In light of these challenges during the creation of this study, the author team elected to use Brandwatch (https://www.brandwatch.com/) to collect social media data. Using this software, data was collected containing the term “student loan” and the hashtag #studentloans from October 1 to October 31, 2023 (the query was exported on November 1, 2023). The query was set to collect all mentions of student loans across various content sources, including X, Reddit, Tumbler, YouTube, news sites, blogs, forums, and review sites. Instead of collecting all data across these platforms, Brandwatch collected a statistically significant sample size with a calculated rate of 33.36%. The export filtered out pornographic content and profanities. This larger dataset (n = 317,394 with 208,595 unique authors) was filtered to collect only records from X (n = 248,303 with 176,640 unique authors), which represented 78.2% of the comprehensive dataset. The final sample for this study was 248,303. A textual and sentiment analysis of the collected data was conducted within the Brandwatch online software. Brandwatch defines sentiment analysis as “the process used to determine the attitude, opinion and emotion expressed by a person about a particular topic in an online mention” (Brandwatch, 2023). A sentiment score is calculated based on the proximity of sentiment and search terms, within the context of larger blocks of text (Brandwatch, 2023). Importantly, if a collection of text cannot be accurately categorized as positive or negative, it is classified as neutral (Brandwatch, 2023). Brandwatch conducts sentiment analysis and calculates sentiment scores using proprietary algorithms, machine learning, and Natural Language Processing (“NLP”) (Brandwatch, 2023). Specifically, Brandwatch uses a pre- trained model using transformer-based deep learning technology to assign sentiment scores (Brandwatch, n.d.-a; Brandwatch, n.d.-b). Transformer models are commonly used to analyze sentiment for social media data and have been used to analyze data from X (Bokolo & Liu, 2024; Gong et al., 2022; Kokab et al., 2022; Padmalal et al., 2024; Sharma et al., 2022). The sentiment model was trained on user annotated sentiment material from the Brandwatch platform, third party datasets from academic studies, annotation service data, and Brandwatch in-house data (Brandwatch, n.d.-a). Sentiment accuracy across languages is estimated to be between 60-75% with an average F1 predictive performance of 55-65% (Brandwatch, n.d.-a). This accuracy range reflects the challenge of conducting a sentiment analysis across a large dataset, specifically, the tradeoff inherent in a data science model between accuracy and bias (Brandwatch, n.d.-a). Brandwatch’s model has been benchmarked to others such as MonkeyLearn, Aylien, Idol, Metamind, AlchemyAPI, and Datumbox with comparable performance (Brandwatch, n.d.-a). Results Figure 1 shows the number of mentions per day across the date range. The mean was 8,009 and https://www.brandwatch.com/ Anderson et al. 41 the median 7,147. Between September 30 and October 23, mentions spiked on four days: October 1 (n = 12,888), October 4 (n = 23,849), October 18 (n = 12,732), and October 31 (n = 14,882). Figure 2 Provides the overall sentiment for student loan mentions within the study’s data range. Most of the mentions were neutral (53%), with 46% negative and 1% positive. This finding supports the first hypothesis, as negative sentiment overwhelmingly outweighs positive sentiment during the date range. Figure 1. Student Loan Mentions by Date Note: n = 248,291. Brandwatch excluded twelve records when completing this analysis. Figure 2. Overall Sentiment for Student Loan Mentions Note: n = 248,301. Brandwatch excluded two records when completing this analysis. 0 5000 10000 15000 20000 25000 30000 1 -O ct 2 -O ct 3 -O ct 4 -O ct 5 -O ct 6 -O ct 7 -O ct 8 -O ct 9 -O ct 1 0 -O ct 1 1 -O ct 1 2 -O ct 1 3 -O ct 1 4 -O ct 1 5 -O ct 1 6 -O ct 1 7 -O ct 1 8 -O ct 1 9 -O ct 2 0 -O ct 2 1 -O ct 2 2 -O ct 2 3 -O ct 2 4 -O ct 2 5 -O ct 2 6 -O ct 2 7 -O ct 2 8 -O ct 2 9 -O ct 3 0 -O ct 3 1 -O ct 1% 53% 46% positive neutral negative Financial Services Review, 33(2) 42 Table 1 provides additional detail on the twenty most popular phrases, persons, and keywords with sentiments of neutral, negative, and positive along with the category sentiment score. The top- ranking keyword was debt, with 106,605 mentions, followed by forgiveness (42,010), billion (32,453), tax (30,804), and money (27,854). The top-ranking person – in fact, the only person listed in the top twenty topics – was Biden (45,422), who also ranked as the second most popular topic across types (President is also listed as twentieth on this list with 17,805 mentions). The top-ranking topics for negative sentiment were debt (69,286 negative sentiment versus 722 positive sentiment), money (24,742 negative sentiment versus 104 positive sentiment), tax (22,976 negative sentiment versus 38 positive sentiment), Biden (20,935 negative sentiment versus 218 positive sentiment), and genocide (20,539 negative sentiment versus 26 positive sentiment). Every topic, keyword, and phrase listed in the top twenty most popular topics had a negative sentiment score, supporting our second hypothesis. To further investigate the findings presented in Table 1 and provide contextual meaning, a positive to negative sentiment ratio was created. This ratio was then compared to the number of mentions. The correlation between the positive- negative sentiment ratio and mentions was calculated to be 0.149, indicating a weak, positive relationship. This weak relationship further supports the findings and gives context to any potential bias in interpretation of high frequency words. When Table 1 was sorted by highest positive-negative sentiment ratio, the top five results form a cluster around student loans: borrowers (0.028), forgiveness (0.026), student loan payments (0.023), payments (0.020), and President (0.015). However, each of these ratios remains quite low at <0.03. Table 2 groups each of the top topics into three broad categories (student loans, government, and other/ambiguous) to better focus on borrower sentiment attached to student loans. The average positive-negative sentiment ratio was calculated for each category. The student loans category had the highest average positive-negative sentiment ratio at 0.021. The next highest categories were government (0.005) and other/ambiguous (0.002). Financial Services Review, 33(2) 43 Table 1. Top 20 Topics by Mentions and Corresponding Sentiment Statistics Topic Name Type Mentions Negative Neutral Positive Sentiment Score +/- Ratio debt Keyword 106605 69286 36596 722 -64 0.010 Biden Person 45422 20935 24268 218 -45 0.010 forgiveness Keyword 42010 16435 25150 425 -38 0.026 billion Keyword 32453 14024 18377 50 -43 0.004 tax Keyword 30804 22976 7788 38 -74 0.002 money Keyword 27854 24742 3006 104 -88 0.004 payments Keyword 24643 10852 13571 218 -43 0.020 relief Keyword 22649 15013 7488 146 -65 0.010 women Keyword 20785 17928 2851 5 -86 0.000 genocide Keyword 20638 20539 71 26 -99 0.001 Americans Keyword 20536 9218 11248 68 -44 0.007 borrowers Keyword 19924 7300 12420 203 -35 0.028 student loan payments Phrase 19906 8613 11095 197 -42 0.023 tax money Phrase 19834 19756 77 0 -99 0.000 children Keyword 19762 18614 1124 23 -94 0.001 voted Keyword 18593 6397 12165 29 -34 0.005 hard Keyword 18557 18029 515 11 -97 0.001 universal Keyword 18200 17697 494 8 -97 0.000 health care Phrase 17883 17754 119 8 -99 0.000 President Keyword 17805 5840 11875 89 -32 0.015 Financial Services Review, 33(2) 44 Table 2. Average Topic Positive-Negative Sentiment Ratio Across Categories Topic Name Type Mentions Negative Neutral Positive Sentiment Score +/- Ratio Category: Student Loans debt Keyword 106605 69286 36596 722 -64 0.010 forgiveness Keyword 42010 16435 25150 425 -38 0.026 payments Keyword 24643 10852 13571 218 -43 0.020 borrowers Keyword 19924 7300 12420 203 -35 0.028 student loan payments Phrase 19906 8613 11095 197 -42 0.023 Average +/- Score 0.021 Category: Government Biden Person 45422 20935 24268 218 -45 0.010 tax Keyword 30804 22976 7788 38 -74 0.002 money Keyword 27854 24742 3006 104 -88 0.004 Americans Keyword 20536 9218 11248 68 -44 0.007 tax money Phrase 19834 19756 77 0 -99 0.000 voted Keyword 18593 6397 12165 29 -34 0.005 health care Phrase 17883 17754 119 8 -99 0.000 President Keyword 17805 5840 11875 89 -32 0.015 Average +/- Score 0.005 Category: Other/Ambiguous billion Keyword 32453 14024 18377 50 -43 0.004 relief Keyword 22649 15013 7488 146 -65 0.010 women Keyword 20785 17928 2851 5 -86 0.000 genocide Keyword 20638 20539 71 26 -99 0.001 children Keyword 19762 18614 1124 23 -94 0.001 hard Keyword 18557 18029 515 11 -97 0.001 universal Keyword 18200 17697 494 8 -97 0.000 Average +/- Score 0.002 Anderson et al. 45 Table 3 displays the twenty most popular phrases with sentiments of neutral, negative, and positive, along with the category sentiment score. The top- ranking phrases were student loan payments (19,906 mentions), followed by tax money (19,834), health care (17,883), forgiving student loan debt (17,469), and funding a genocide (17,217). Regarding the mix of sentiment, 16/20 (80%) of topics had a negative sentiment score, and 4/20 (20%) had a neutral sentiment score of 0. The top-ranking topics for positive sentiment were student loan payments (197 positive sentiment versus 8,613 negative sentiment), student loan borrowers (38 positive sentiment versus 4,583 negative sentiment), Supreme Court (26 positive sentiment versus 11,908 negative sentiment), student debt relief (17 positive sentiment versus 8,154 negative sentiment), and voted for student loan forgiveness (17 positive sentiment versus 119 negative sentiment). Even though they ranked highest for positive sentiment, each of these phrases had a negative overall sentiment score. No topics listed had a positive overall sentiment score. In fact, four topics on the list - tax money, health care, forgiving student loan debt, and funding a genocide - had the maximum negative sentiment score of -99. This finding offers additional support for our second hypothesis. A positive-negative sentiment ratio was created for phrases to compare the relationship between positive and negative sentiment. This ratio was then compared to the number of mentions. The correlation between the positive-negative sentiment ratio and mentions was calculated to be -0.078 (when rows with score of 0 for both positive and negative sentiment were assigned a ratio score of 0) or -0.122 (when rows with score of 0 for both positive and negative sentiment were excluded from the analysis). These correlations indicate a weak, negative relationship between the positive-negative sentiment ratio and mentions. When Table 3 was sorted by highest positive-negative sentiment ratio, the top five results were voted for student loan forgiveness (0.143), student loan payments (0.023), student loan borrowers (0.008), millions of Americans (0.004), and billion in student loan debt (0.003). Each of these phrases had a positive-negative sentiment ratio of <0.15. The eleventh most popular phrase #fitness Check comments for full video was unrelated to the topic of student loans. Similarly, the seventeenth most popular phrase, bringing him home pt3 #fitness, was also unrelated to student loans. The topic of fitness—and why it might have shown up in the results—is expounded in subsequent parts of the paper. Table 4 groups the popular phrases into the previously utilized three categories of student loans, government, and other/ambiguous. The average positive-negative sentiment ratio was calculated for each category. The student loans category had the highest average positive- negative sentiment ratio at 0.02 with the other two categories (other/ambiguous and government) tied with 0.001. Financial Services Review, 33(2) 46 Table 3. Top 20 Phrases and Corresponding Sentiment Statistics Rank Topic Name Mentions Negative Neutral Positive Sentiment Score +/- Ratio 1 student loan payments 19906 8613 11095 197 -42 0.023 2 tax money 19834 19756 77 0 -99 0.000 3 health care 17883 17754 119 8 -99 0.000 4 forgiving student loan debt 17469 17427 35 5 -99 0.000 5 funding a genocide 17217 17214 2 0 -99 0.000 6 Supreme Court 13832 11908 1897 26 -85 0.002 7 student loan billing statement 12165 2092 10070 2 -17 0.001 8 Helping 18yo 11266 0 11266 0 0 0.000 9 student loan borrowers 8925 4583 4302 38 -50 0.008 10 student debt relief 8877 8154 704 17 -91 0.002 11 #fitness Check comments for full video 8811 0 8811 0 0 0.000 12 million Americans 8796 6553 2233 8 -74 0.001 13 voted for student loan forgiveness 8607 119 8469 17 -1 0.143 14 billion in student loan debt 8163 2986 5168 8 -36 0.003 15 student with his student loan 8139 0 8139 0 0 0.000 16 voted to restart 8016 4625 3390 0 -57 0.000 17 bringing him home pt3 #fitness 7638 0 7638 0 0 0.000 18 cancel your student debt 7192 3945 3246 0 -54 0.000 19 canceling an additional 6196 2305 3891 0 -37 0.000 20 millions of Americans 5597 470 5123 2 -8 0.004 Anderson et al. 47 Table 4. Average Phrase Positive-Negative Sentiment Ratio Across Categories Topic Name Mentions Negative Neutral Positive Sentiment Score +/- Ratio Category: Student Loans student loan payments 19906 8613 11095 197 -42 0.023 forgiving student loan debt 17469 17427 35 5 -99 0.000 student loan billing statement 12165 2092 10070 2 -17 0.001 student loan borrowers 8925 4583 4302 38 -50 0.008 student debt relief 8877 8154 704 17 -91 0.002 voted for student loan forgiveness 8607 119 8469 17 -1 0.143 billion in student loan debt 8163 2986 5168 8 -36 0.003 student with his student loan 8139 0 8139 0 0 0.000 cancel your student debt 7192 3945 3246 0 -54 0.000 Average +/- Score 0.020 Category: Government tax money 19834 19756 77 0 -99 0.000 health care 17883 17754 119 8 -99 0.000 Supreme Court 13832 11908 1897 26 -85 0.002 Average +/- Score 0.001 Category: Other/Ambiguous funding a genocide 17217 17214 2 0 -99 0.000 Helping 18yo 11266 0 11266 0 0 0.000 #fitness Check comments for full video 8811 0 8811 0 0 0.000 million Americans 8796 6553 2233 8 -74 0.001 voted to restart 8016 4625 3390 0 -57 0.000 bringing him home pt3 #fitness 7638 0 7638 0 0 0.000 canceling an additional 6196 2305 3891 0 -37 0.000 millions of Americans 5597 470 5123 2 -8 0.004 Average +/- Score 0.001 Financial Services Review, 33(2) 48 Table 5 shows the top emojis used across all tweets gathered. Although this table is not directly related to a hypothesis – or attached to sentiment scores – it gives helpful insights into the mindset of the authors when writing about student loan topics. Across all tweets (tweets and retweets), the five most popular emojis were smiling face with horns (n = 11,704), money bag (n = 1,864), police cars revolving light (n = 1,696), spool of thread (n = 1,202), and white down pointing backhand index (n = 1,067). For tweets, the five most popular emojis were money bag (n = 595), face with tears of joy (n = 490), loudly crying face (n = 310), white down pointing backhand index (n = 94), and police cars revolving light (n = 55). When sorted by the greatest total number of impressions, the five most popular emojis were smiling face with horns (384,182,044), white down pointing backhand index (102,521,059), money bag (59,243,121), splashing sweat symbol (39,698,047), and movie camera (36,811,251). Of the ten top emojis, at least three carry a negative connotation (smiling face with horns, police cars revolving light, loudly crying face) with 14,068 total tweets and retweets and 392,030,725 impressions. Table 5. Top 10 Emojis Rank Emoji Label All Tweets Retweets Tweets Impressions 1 😈 smiling face with horns 11,704 11,701 2 384,182,044 2 💰 money bag 1,864 1,268 595 59,243,121 3 🚨 police cars revolving light 1,696 1,639 55 6,804,599 4 🧵 spool of thread 1,202 1,154 46 4,542,805 5 👇 white down pointing backhand index 1,067 971 94 102,521,059 6 🎥 movie camera 1,052 1,049 2 36,811,251 7 💦 splashing sweat symbol 1,001 992 8 39,698,047 8 😂 face with tears of joy 743 251 490 11,401,407 9 🤷🏽‍♂️ shrugging 734 719 14 1,468,062 10 😭 loudly crying face 668 356 310 1,044,082 Table 6 shows the top hashtags for the data gathered. Like emojis, these hashtags are not attached to sentiment scores. The top hashtag is unrelated to student loans: #fitness. Interestingly, this hashtag had 12,753 retweets but no tweets. Beyond this anomaly (addressed in the limitations section), the second to tenth hashtags were all related to student loans. The second most popular hashtag was #studentloans followed by #cancelstudentdebt. The sixth most popular hashtag was #scotus, referring to the Supreme Court of the United States. For tweets (versus retweets), the five most popular hashtags were #studentloans (1,984), #cancelstudentdebt (775), #poortax (386), #studentloan (262), and #cancelallstudentdebtnow (235). The top five hashtags had combined impressions of 34,533,799, with the list garnering 537,460,898. Anderson et al. 49 Table 6. Top 10 Hashtags Rank Hashtag All Tweets Retweets Tweets Impressions 1 #fitness 12,753 12,753 0 501,162,722 2 #studentloans 3,567 1,582 1,984 23,392,392 3 #cancelstudentdebt 1,933 1,157 775 3,600,657 4 #studentloan 674 410 262 7,267,211 5 #poortax 446 59 386 146,290 6 #scotus 428 395 31 646,113 7 #fixstudentloans 404 368 34 342,014 8 #studentloanforgiveness 305 122 181 650,748 9 #cancelallstudentdebtnow 266 29 235 127,249 10 #cancelallstudentdebt 260 179 79 125,502 Discussion The years and months preceding October 2023 were filled with a whirlwind of change in the federal student loan space. On March 13, 2020, the Trump Administration announced the suspension of federal student loan payments and interest. A few years later, on August 24, 2022, the Biden Administration announced blanket student loan forgiveness while payments and interest were still suspended. On June 30 of the following year the Supreme Court rejected that forgiveness proposal. In the wake of this announcement, borrowers were told that payments and interest would resume – without any balance forgiven – after over three years of no payments or interest accrual on federal student loans. This study’s analysis of X data advances a better understanding of student loan borrower sentiment during the resumption of federal student loan payments in October 2023. Our study demonstrates that borrowers experienced negative sentiment during this resumption of payments, especially without the help of forgiveness. References to student loans on the platform demonstrated a negative sentiment that greatly outpaced positive (46% negative versus 1% positive). Given the timeline outlined above, it is unsurprising that several of the most popular hashtags found in this study pointed toward student loan forgiveness as a major topic, as shown in the third, sixth, eighth, ninth, and tenth most popular hashtags (#cancelstudentdebt, #scotus, #studentloanforgiveness, #cancelallstudentdebtnow, and #cancelallstudentdebt, respectively). In summary, both of our hypotheses were confirmed. Although neutral sentiment represented the greatest percentage across all data points (53%), negative sentiment considerably outweighed positive during the resumption of federal student loan payments. Furthermore, the topics and phrases with negative sentiment ranked higher in mentions than those with positive or neutral sentiment. All the top-ranking topics had negative sentiment scores, while phrases had higher-ranking negative sentiment scores (with negative sentiment scores taking the top seven slots and 80% of the top twenty list). Although student loan topics as a category ranked higher in average positive-negative sentiment ratios, the correlations between positive-negative sentiment ratios and mentions remained weak, pointing to a lack of bias in the findings. Implications Financial planners interacting with clients after the payment resumption should introduce appropriate resources to help borrowers cope with the negative feelings they may experience surrounding this area of their financial lives. Luckily for the profession, this study’s findings align perfectly with the emergence of financial therapy as a bona fide discipline within the field of personal financial planning. Many practitioners in this niche, associations such as Financial Services Review, 33(2) 50 the Financial Therapy Association (FTA), and personal financial planning academic programs such as the one at Kansas State University, offer training and resources for financial planners hoping to support clients in distressing financial situations. Similarly, the CFP Board has put greater emphasis on teaching the behavioral aspects of financial planning with the addition of the “psychology of financial planning” in the CFP® Certification 2021 Principal Knowledge Topics list (CFP® Certification 2021 Principal Knowledge Topics 2021). This means the topic is now taught in all registered educational programs and tested on the CFP® exam. Whatever the preferred term (i.e., financial therapy or behavioral finance), teaching planners psychological support mechanisms has significant potential to benefit financial planning clients and student loan borrowers. Practitioners would be wise to heed the warnings from industry thought-leaders on how student loans can delay a client’s financial growth. Previously outlined studies, such as the one from Archuleta et al. (2013), demonstrate the negative aspects of carrying student loan debt, including the alarming connection between student loans and financial anxiety. In a recent article for Business Insights, researcher and financial therapist Dr. Megan McCoy explained how student loans can harm borrowers in two crucial ways: increasing shame and delaying financial milestones (Aguino & Richtmyer, 2023). Both can be devastating for a client’s progress toward financial independence. The mounting evidence is a call to action for advisors; once and for all, student loans deserve a dedicated space in the financial plan. The costs are too great – and only increasing – for the emerging generation already saddled with student loan debt. Limitations This study does not compare sentiment during repayment to a period beforehand. As such, no comparisons can be made between how sentiment changed before the federal student loan interest and payment restart and after. Without the ability to compare to previous sentiment, it is impossible to decipher if negative sentiment is reflective of student loans (in general), the resumption of federal student loan payments and interest, or the collapse of Biden’s student loan forgiveness initiative. Additionally, textual patterns and sentiment captured during the month of October might not be fully reflective of when student loan payments resumed as a whole, as borrowers had different due dates within the month. This research project was largely descriptive in nature, using frequency-based analysis as the main way to uncover patterns in these gathered data. This tool was useful in identifying the overall narrative within the dataset, which can be especially helpful for exploratory research projects. However, relying too much on frequency-based analysis when using NLP can, at times, influence the interpretation of the results, introduce bias, or ignore broader context. As such, future research on this topic should use tools like regression analysis to push further into the investigation of concrete associations and relationships. Accordingly, this paper’s conclusions should be interpreted in concert with those future contributions. The results displayed in Table 6 showed that #fitness was the top hashtag across our dataset. Table 3 also showed that the phrase “#fitness Check comments for full video” was the eleventh-ranking phrase in the dataset. Unfortunately, social media data sources contain noise which can cloud the signal. Given the phrase had a sentiment score of 0, it is unlikely this result affected the testing of the study’s first hypothesis, although it is probable the phrase artificially boosted overall neutral sentiment. To gain a better understanding of the fitness phenomenon, on October 24, 2023, the authors searched the hashtags #studentloans and #fitness directly on X’s website. A review of the results showed that these two hashtags regularly appeared together within large groupings of hashtags (>5). It is possible the hashtag #fitness was used along with #studentloans to boost search ranking, but perhaps more research could uncover the reasoning behind this peculiar finding. As mentioned previously, the Brandwatch query was set to filter out pornographic content and profanities. This filtering of profanities may have artificially lowered negative sentiment for this study, even though it was notably higher than Anderson et al. 51 positive sentiment already. When the initial query was modified to filter out these categories, the query generator indicated this filter decreased the number of collected responses by as much as 50%.6 If included, this data would have likely pushed negative sentiment even higher. Conclusions This study used Brandwatch to analyze borrower sentiment on the X platform during the resumption of federal student loan payments starting in the month of October 2023. The emotions captured by this data-gathering and analysis process were quite bleak; many borrowers expressed overwhelmingly negative emotions directed at this change. However, this negative outlook highlights an opportunity for financial planners to proactively provide adequate support for the financial and mental health of federal student loan borrowers. Given appropriate and timely action, even negative student loan events can further solidify the financial planners’ positive presence in their clients’ lives. References Aguino, L. & Richard Richtmyer, R. 2023. “Student loans aren't just bad for your wallet — they're bad for your mental health, too.” Business Insider. Accessed April 25, 2024, https://www.businessinsider.com/person al-finance/student-loans-negatively- affect-borrowers-mental-health-2022-9. Archuleta, K. L., Dale, A., & Spann, S. M. (2013). College students and financial distress: Exploring debt, financial satisfaction, and financial anxiety. Journal of Financial Counseling and Planning, 24(2), 50–62. Bokolo, B. G., & Liu, Q. (2024). Advanced comparative analysis of machine learning and transformer models for depression and suicide detection in social media texts. Electronics, 13(20), 3980. https://doi.org/10.3390/electronics13203 6 A preliminary review of the data before filtering showed a large number of records containing the f- word 980 Brandwatch. (n.d.-a). BCR sentiment methodology and evaluation. Brandwatch. (n.d.-b). Sentiment analysis – consumer research. Brandwatch Help. Retrieved October 30, 2024, from https://consumer-research- help.brandwatch.com/hc/en- us/restricted?return_to=https%3A%2F% 2Fconsumer-research- help.brandwatch.com%2Fhc%2Fen- us%2Farticles%2F360013739958- Sentiment-Analysis Brandwatch (2023, September 20). Sentiment analysis. https://www.brandwatch.com/wp- content/uploads/brandwatch/Sentiment- Analysis.pdf Carvalho, J., & Plastino, A. (2021). On the evaluation and combination of state-of- the-art features in Twitter sentiment analysis. Artificial Intelligence Review, 54(3), 1887–1936. https://doi.org/10.1007/s10462-020- 09895-6 CFP® Certification 2021 Principal Knowledge Topics (2021). Accessed April 25, 2024. https://www.cfp.net/-/media/files/cfp- board/cfp-certification/2021-practice- analysis/2021-principal-knowledge- topics.pdf. Chancellor, S., & De Choudhury, M. (2020). Methods in predictive techniques for mental health status on social media: A critical review. Npj Digital Medicine, 3(1), Article 1. https://doi.org/10.1038/s41746-020- 0233-7 Financial Services Review, 33(2) 52 Coppersmith, G., Dredze, M., & Harman, C. (2014). Quantifying mental health signals in Twitter. Proceedings of the Workshop on Computational Linguistics and Clinical Psychology: From Linguistic Signal to Clinical Reality, 51– 60. https://doi.org/10.3115/v1/W14- 3207 Coppersmith, G., Harman, C., & Dredze, M. (2014). Measuring Post Traumatic Stress Disorder in Twitter. Proceedings of the International AAAI Conference on Web and Social Media, 8(1), Article 1. https://doi.org/10.1609/icwsm.v8i1.1457 4 Coppersmith, G., Leary, R., Crutchley, P., & Fine, A. (2018). Natural Language Processing of social media as screening for suicide risk. Biomedical Informatics Insights, 10, 1178222618792860. https://doi.org/10.1177/11782226187928 60 De Choudhury, M., Gamon, M., Counts, S., and Horvitz, E. (2013). Predicting depression via social media. Proceedings of the International AAAI Conference on Web and Social Media 7(1): 128–37. https://doi.org/10.1609/icwsm.v7i1.1443 2 Ebrahimian, M. (2023). Student Loans and Social Mobility (SSRN Scholarly Paper 3680159). https://doi.org/10.2139/ssrn.3680159 Edo-Osagie, O., De La Iglesia, B., Lake, I., & Edeghere, O. (2020). A scoping review of the use of Twitter for public health research. Computers in Biology and Medicine, 122, 103770. https://doi.org/10.1016/j.compbiomed.2 020.103770 Elliott, W., Grinstein-Weiss, M., & Nam, I. (2013). “Student debt and declining retirement savings (Working Paper No. 13‐34).” Center for Social Development. Fry, R. (2012). A record one-in-five households now owe student loan debt. Pew Research Center. Giachanou, A., & Crestani, F. (2017). Like it or not: A survey of Twitter sentiment analysis methods. ACM Computing Surveys, 49(2), 1–41. https://doi.org/10.1145/2938640 Gicheva, D., & Thompson, J. (2014). The effects of student loans on long-term household financial stability. Department of Economics Working Paper no. 14-02. Greensboro: University of North Carolina. Gong, X., Ying, W., Zhong, S., & Gong, S. (2022). Text sentiment analysis based on transformer and augmentation. Frontiers in Psychology, 13, 906061. https://doi.org/10.3389/fpsyg.2022.9060 61 Guntuku, S. C., Preotiuc-Pietro, D., Eichstaedt, J. C., & Ungar, L. H. (2019). What Twitter profile and posted images reveal about depression and anxiety. Proceedings of the International AAAI Conference on Web and Social Media, 13, 236–246. https://doi.org/10.1609/icwsm.v13i01.32 25 Guntuku, S. C., Yaden, D. B., Kern, M. L., Ungar, L. H., & Eichstaedt, J. C. (2017). Detecting depression and mental illness on social media: An integrative review. Current Opinion in Behavioral Sciences, 18, 43–49. https://doi.org/10.1016/j.cobeha.2017.07 .005 Hanson, M (2023, September 20). Economic Effects of Student Loan Debt. EducationData.org. https://educationdata.org/student-loan- debt-economic-impact Joseph, M., & MacDonald, M. (2021). Investigating antecedents to financial satisfaction in emerging adults. 2021 Academic Research Colloquium. https://www.researchgate.net/publicatio n/355903050_Investigating_Antecedent s_to_Financial_Satisfaction_in_Emergin g_Adults https://doi.org/10.1609/icwsm.v13i01.3225 https://doi.org/10.1609/icwsm.v13i01.3225 https://doi.org/10.1016/j.cobeha.2017.07.005 https://doi.org/10.1016/j.cobeha.2017.07.005 Anderson et al. 53 Kim, J., & Chatterjee, S. (2019). Student loans, health, and life satisfaction of US households: Evidence from a panel study. Journal of Family and Economic Issues, 40(1), 36–50. https://doi.org/10.1007/s10834-018- 9594-3 Kim, K. T., Lee, J. M., & Lee, J. (2021). Student loans and financial satisfaction: The moderating role of financial education. Journal of Financial Counseling and Planning. https://doi.org/10.1891/JFCP- 19-00002 Kokab, S.T., Asghar, S., & Naz, S. (2022). Transformer-based deep learning models for the sentiment analysis of social media data. Array, 14, 100157. https://doi.org/10.1016/j.array.2022.100 157 Kouloumpis, E., Wilson, T., & Moore, J. (2011). Twitter sentiment analysis: The good the bad and the OMG! Proceedings of the International AAAI Conference on Web and Social Media, 5(1), Article 1. https://doi.org/10.1609/icwsm.v5i1.1418 5 Mountain, T. P., Cao, X., Kim, N., & Gutter, M. S. (2020). Millennials’ future homeownership and the role of student loan debt. Family and Consumer Sciences Research Journal, 49(1), 5–23. https://doi.org/10.1111/fcsr.12374 Nadeem, M., Horn, M., Coppersmith, G., & Sen, S. (2016). Identifying depression on Twitter (arXiv:1607.07384). arXiv. https://doi.org/10.48550/arXiv.1607.073 84 National Center for Education Statistics. (2023, November 11). Price of attending an undergraduate institution. U.S. Department of Education, Institute of Education Sciences. https://nces.ed.gov/programs/coe/indicat or/cua Nuckols, W., Bullington, K. E., & Gregory, D. E. (2020). Was it worth it? Using student loans to finance a college degree. Higher Education Politics & Economics, 6(1), Article 1. https://doi.org/10.32674/hepe.v6i1.1358 Ortiz-Ospina, E. (2019, September 20). The rise of social media. OurWorldInData.org. https://ourworldindata.org/rise-of-social- media Padmalal, S., Dayanand, I. E., Rao, G. S., & Gore, S. (2024). Enhancing sentiment analysis in social media texts using Transformer- Based NLP Models. International Journal of Electrical and Electronics Engineering, 11(8), 208–216. https://doi.org/10.14445/23488379/IJEE E-V11I8P118 Preoţiuc-Pietro, D., Eichstaedt, J., Park, G., Sap, M., Smith, L., Tobolsky, V., Schwartz, H. A., & Ungar, L. (2015). The role of personality, age, and gender in tweeting about mental illness. Proceedings of the 2nd Workshop on Computational Linguistics and Clinical Psychology: From Linguistic Signal to Clinical Reality, 21–30. https://doi.org/10.3115/v1/W15-1203 Qian, Y., & Fan, W. (2023). Student loans, mental health, and substance use: A gender comparison among US young adults. Journal of American College Health, 71(3), 930–941. https://doi.org/10.1080/07448481.2021. 1909046 Reece, A. G., Reagan, A. J., Lix, K. L. M., Dodds, P. S., Danforth, C. M., & Langer, E. J. (2017). Forecasting the onset and course of mental illness with Twitter data. Scientific Reports, 7(1), 13006. https://doi.org/10.1038/s41598-017- 12961-9 Robb, C. A., Chatterjee, S., Porto, N., & Cude, B. J. (2019). The influence of student loan debt on financial satisfaction. Journal of https://doi.org/10.48550/arXiv.1607.07384 https://doi.org/10.48550/arXiv.1607.07384 https://doi.org/10.1038/s41598-017-12961-9 https://doi.org/10.1038/s41598-017-12961-9 Financial Services Review, 33(2) 54 Family and Economic Issues, 40(1), 51– 73. https://doi.org/10.1007/s10834-018- 9599-y Sato, Y., Watt, R. G., Saijo, Y., Yoshioka, E., & Osaka, K. (2020). Student loans and psychological distress: A cross-sectional study of young adults in Japan. Journal of Epidemiology, 30(10), 436–441. https://doi.org/10.2188/jea.JE20190057 Sharma, U., Pandey, P., & Kumar, S. (2022). A transformer-based model for evaluation of information relevance in online social- media: A case study of Covid-19 media posts. New Generation Computing, 40(4), 1029–1052. https://doi.org/10.1007/s00354-021- 00151-1 Sinha, G. R., Larrison, C. R., Brooks, I., & Kursuncu, U. (2023). Comparing naturalistic mental health expressions on student loan debts using Reddit and Twitter. Journal of Evidence-Based Social Work, 20(5), 727–742. https://doi.org/10.1080/26408066.2023. 2202668 Tsugawa, S., Kikuchi, Y., Kishino, F., Nakajima, K., Itoh, Y., & Ohsaki, H. (2015). Recognizing depression from Twitter activity. Proceedings of the 33rd Annual ACM Conference on Human Factors in Computing Systems, 3187–3196. https://doi.org/10.1145/2702123.270228 0 Zhang, Q., & Kim, H. (2019). American young adults’ debt and psychological distress. Journal of Family and Economic Issues, 40(1), 22–35. https://doi.org/10.1007/s10834-018- 9605-4 Zimbra, D., Abbasi, A., Zeng, D., & Chen, H. (2018). The state-of-the-art in Twitter sentiment analysis: A review and benchmark evaluation. ACM Transactions on Management Information Systems, 9(2), 1–29. https://doi.org/10.1145/3185045