Landing page

Introduction

In our daily life, we often encounter such a situation: when we try to give feedback on problems to enterprises or institutions, we always struggle with which category our appeal belongs to; And when we try to find other people’s feedback on a product or specific issue, we often get the wrong answer. This kind of classification confusion not only disturbs users, but also affects the efficiency of enterprises or institutions to deal with feedback to a certain extent. Because customers are often unable to accurately categorize their demands, the staff receiving feedback may not be able to respond quickly and effectively, and ultimately the problem is not properly solved.

We initially rise these questions: what’s the relationship between each information data of complaint? and which information is most dependent?

For this project, we obtained customer complaint data from the CFPB (Consumer Financial Protection Bureau) during period from April 1, 2017 to August 31, 2023. We believe that the most true reflection of customer demands is the content of the complaint described by the customer himself, regardless of what the complaints are categorized as. Thus, we rised questions that how much does the categories actually related to complaints content?

To figure this out, we used the data set combining complaint texts and other information about product categories and complaint issues to conduct an in-depth analysis of user complaints. The study found that the existing website classification system (such as product and issue category) has a large deviation from the actual complaint content of users, and can not accurately reflect the real demands of customers. Then we came up with an idea: is it feasible to categorize complaints simply base on the complaint text?

In order to solve this problem, we try to input the user’s processed complaint text data into a clustering model such as K-Means for automatic classification. The results show that text clustering directly based on user complaint content is a feasible path, which can reveal the intrinsic relationship between complaint content more naturally. In the future, we will further analyze the clustering results and define clear topics for each type of complaint content.

Another idea came into our mind. Beside get to know what’s user’s request, knowing user’s attitude is also important. So we wondered can we obtain user’s attitude simply base on what kind of product with which kind of issue is bothering them? This may help to prioritize the processing order. To find this out, we dived into the relationship between categories and complaint sentiment.

This work will help enterprises and organizations optimize the feedback processing process, so that customer demands can be more accurately matched to the corresponding departments, and improve the efficiency of problem solving and customer satisfaction.

Through this project, we hope to provide enterprises and organizations with a new clue of classification methods based on real user needs, improve the shortcomings of the current classification system, and build a more efficient and intelligent feedback processing system.

Literature review

Qingyang Wang

  1. Smart Complaint Management System1
    The Smart Complaint Management System (SCMS) addresses customer dissatisfaction by providing a mobile app, chatbot, and web application for filing complaints. The system classifies complaints, directs them to the responsible department, and identifies similar complaints to avoid duplication. Test results show it reduces complaint handling time, improves filing channels, and enhances progress tracking.

  2. Financial Complaints: Sentiment Analysis: Final Technical Report2
    This research proposes an explainable complaint cause identification system using a dyadic attention mechanism to classify complaints, detect sentiment, and recognize emotions. It adds causal span annotations to a financial complaints corpus, aiming to improve complaint detection, severity classification, and identify causes, offering new opportunities for research in natural language processing.

  3. Complaint and severity identification from online financial content3
    This article introduces FINCORP, a resource of annotated financial complaints expressed on Twitter, enriched with emotion, sentiment, and severity classifications. It develops a multitask framework for complaint detection, severity classification, emotion recognition, and sentiment classification, and compares its performance with existing baselines to evaluate the dataset.

  4. Negative Review or Complaint? Exploring Interpretability in Financial Complaints4
    This paper proposes an explainable complaint cause identification approach using a dyadic attention mechanism at the sentence and word levels to classify financial complaints. The model simultaneously trains tasks for complaint detection, sentiment detection, and emotion recognition. The study adds causal span annotations to an existing financial complaints corpus and highlights the potential of conventional computing techniques for solving new problems.

  5. Data analysis of consumer complaints in banking industry using hybrid clustering5
    This paper explores Consumer Finance Complaints data to analyze the frequency of similar complaints related to specific banks, services, or products. Using data mining, cluster analysis, and predictive modeling, the study provides insights into complaint patterns, helping banks identify areas of concern, improve customer satisfaction, and enhance profitability.

Binghui Ni

  1. A classification-based approach for modelling disputed responses based on consumer complaint on financial products6
    This study aims to identify the best-performing model for predicting the likelihood of consumers disputing complaints responses by financial service providers. Three models—Naive Bayes, Random Forest, and Logistic Regression—were compared, with Random Forest achieving the highest accuracy. Thus, Random Forest was selected as the optimal model for this task.

  2. Managing complaints to improve customer profitability7
    This study examines how organizational responses to customer complaints (timeliness, compensation, and communication) influence customer profitability. It proposes a contingency framework, tested in the financial services industry, showing that response effectiveness varies based on customer relationship strength and failure type. The results offer insights into tailored complaint-handling strategies to enhance profitability.

  3. Statistical analysis and prediction of the product complaints8
    This article analyzes cardboard packaging complaints using quality and statistical tools to assess corrective and preventive actions’ effectiveness. The study evaluates complaints over one year and quarterly, focusing on customer complaints and financial losses. The analysis identified the critical complaint code 403 (overprint) and predicted future complaints, showing that the company’s actions have not yet reduced the complaint numbers.

  4. Latent Dirichlet allocation (LDA) for topic modeling of the CFPB consumer complaints9
    This paper proposes an intelligent system using latent Dirichlet allocation (LDA) to analyze consumer complaints processed by the Consumer Financial Protection Bureau (CFPB). The system extracts latent topics from complaint narratives and explores their time trends to evaluate the effectiveness of CFPB regulations, aiding experts in improving consumer experience through more efficient investigations.

  5. The financial restitution gap in consumer finance: Insights from complaints filed with the CFPB10
    This study finds that consumers from low-socioeconomic zip codes are 30% less likely to receive financial restitution for their complaints filed with the CFPB. While the socioeconomic gap was minimal under the Obama administration, it widened under the Trump administration, attributed to industry-friendly policies and leadership.

References

1.
Kormpho, P., Liawsomboon, P., Phongoen, N. & Pongpaichet, S. Smart complaint management system. in 2018 seventh ICT international student project conference (ICT-ISPC) 1–6 (2018). doi:10.1109/ICT-ISPC.2018.8523949.
2.
Moleka, A. J. Financial complaints: Sentiment analysis: Final technical report. (Dublin, National College of Ireland, 2017).
3.
Singh, A., Bhatia, R. & Saha, S. Complaint and severity identification from online financial content. IEEE Transactions on Computational Social Systems 11, 660–670 (2023).
4.
Das, S., Singh, A., Saha, S. & Maurya, A. Negative review or complaint? Exploring interpretability in financial complaints. IEEE Transactions on Computational Social Systems 11, 3606–3615 (2024).
5.
Chugani, S., Govinda, K. & Ramasubbareddy, S. Data analysis of consumer complaints in banking industry using hybrid clustering. in 2018 second international conference on computing methodologies and communication (ICCMC) 74–78 (IEEE, 2018).
6.
Onunkwo, C. A classification-based approach for modelling disputed responses based on consumer complaint on financial products. (Dublin, National College of Ireland, 2019).
7.
Cambra-Fierro, J., Melero, I. & Sese, F. J. Managing complaints to improve customer profitability. Journal of Retailing 91, 109–124 (2015).
8.
Knop, K. & Ziora, R. Statistical analysis and prediction of the product complaints. System Safety: Human-Technical Facility-Environment 4, 99–115 (2022).
9.
Bastani, K., Namavari, H. & Shaffer, J. Latent dirichlet allocation (LDA) for topic modeling of the CFPB consumer complaints. Expert Systems with Applications 127, 256–271 (2019).
10.
Haendler, C. & Heimer, R. The financial restitution gap in consumer finance: Insights from complaints filed with the CFPB. Available at SSRN 3766485 (2021).