Increasing the effectiveness of Recommendation System

1. Recommendation Algorithm have been accused to trap users in
a filter bubble, to promote balkanization of information and
hence to reduce serendipity.
2. How can reviews which contain rich information be used for
optimization.
3. How to solve Cold Start Problem.
4. How to solve Echo Chamber effect and Long Tail Effect.

Transcript

1. Increasing effectiveness of Recommendation System TECH-04
2. INTRODUCTION 1. Recommendation Algorithm have been accused to trap users in a filter bubble, to promote balkanization of information and hence to reduce serendipity. 2. How can reviews which contain rich information be used for optimization. 3. How to solve Cold Start Problem. 4. How to solve Echo Chamber effect and Long Tail Effect.
3. AIM & OBJECTIVE The aim is to increase effectiveness of recommendation system by increasing the coverage for every product and personalization for every user. Objective is to learn how recommender systems works and solve the various problems such as cold start, echo chamber, long tail effect etc, making recommendation system better.
4. NEED & SIGNIFICANCE Recommendation System can be used in a variety of applications, such as recommending movies to watch on a streaming platform, products to purchase on e-commerce website, articles to read in a news app, banking, retail, telecom, utilities, etc.
5. HYPOTHESIS : REVIEWS Reviews contains rich information about user’s taste. We convert each word in review text into distributed representation in form of word vector which serves as input to the hidden layer of RNN. The output of model is prediction of probability that a user will like product associated with input review.
6. HYPOTHESIS : COLD START PROBLEM Reasons : 1. Systematic Bootstrapping. 2. Low Interaction 3. New User Solutions : 1. Representative approach 2. Feature Mapping 3. Hybrid Approach 4. Drop-out Net 5. Session-based RNN 6. Explore/Exploit Tradeoff
7. HYPOTHESIS : ECHO CHAMBER Problems : 1. Users Interest are Diverse. 2. Indirect Interaction between users and the platform. Solutions : 1. Reinforcement in User Interests. 2. Change in Content Diversity. 3. Explore/Exploit Tradeoff
8. HYPOTHESIS : LONG TAIL EFFECT Problems : 1. Sparsity. 2. Profitability. 3. Uniqueness. Solutions : 1. Head/Tail Partitioning. 2. Clustering. 3. Explore/Exploit Tradeoff.
9. CONCLUSION Solving all the hypothesis, we increase the effectiveness of the recommendation system. And we can also conclude that recommender systems do not create a filter bubble. Further effectiveness of the recommendation system can be increased by use of Reinforcement Learning, Sentiment Analysis, Feature Engineering etc.