Loading...

Prediction of Customer Churn From Subscription Services in Response to Recommendations: With Emphasis on MCI Data

Shirali, Ali | 2021

672 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 54378 (05)
  4. University: Sharif University of Technology
  5. Department: Electrical Engineering
  6. Advisor(s): Amini, Arash; Kazemi, Reza
  7. Abstract:
  8. In competitive markets where a product or service is provided by multiple providers, as the telecom market, keeping active users is expected to be less expensive than attracting new users. In this regard, first of all, churning should be predicted for active users, and secondly, proper recommendations should be provided to prevent churning. In this thesis, by modeling customer churn as a response to the recommendations, we study the churn prediction and prevention problem as a recommender system. This model enables us to select the best offer for each user to prevent it from churning.Modeling customer churn in a recommender system introduces new challenges, including delay in observing responses, asymmetry of the problem (when the number of users is much larger than the number of possible recommendations), and observing biased ratings. Specifically, delay in observing responses makes data collection challenging. In collaboration with the Mobile Telecommunication Company of Iran (MCI), we collected a new dataset containing 20,000 users and evaluated all proposed solutions.To tackle the introduced challenges, we have proposed two solutions. The first solution is a smooth recommender in the frequency domain. Here, with a new perspective to the problem of rating prediction, users are described with smooth band-limited scoring functions, and the problem of rating prediction is reduced to the problem of signal recovery from non-uniform distributed samples. To solve this new problem, we have proposed two methods; in the first method, we have employed an alternating optimization technique, and in the second method, we have introduced an interpretable feed-forward neural network. In evaluating the performance of the proposed solution, we have shown that this is not only an excellent solution to the asymmetry challenge of the churn prediction problem, but it works better or is close to the state-of-the-art methods on benchmark datasets.In the second proposed solution, we propose a solution to the general problem of the rating prediction by modeling users on a directed weighted graph. Here, users are vertices, and their responses to a recommendation are a signal on this graph. So, the problem of rating prediction is reduced to the graph signal recovery. We show that if we learn graph weights concurrently with recovering signals, the performance will be significantly improved. Based on the evaluations, this solution is effective in churn prediction and works close to the state-of-the-art methods on benchmark datasets.Finally, in the last chapter, we have revised the proposed solutions to make them robust to the biased input. By addressing this last challenge in the two proposed solutions, we have successfully modeled the churn prediction problem in a recommender system and tackled all of its challenges
  9. Keywords:
  10. Customer Churn Prediction ; Recommender System ; Collaborative Filtering ; Representation Learning ; Domain Adaptation ; Graph Signal Recovery

 Digital Object List

 Bookmark