Seminar Title
Design an Efficient Movie Group Recommender System through Various Approaches Using Film Preferences
Seminar Type
Progress Seminar
Department
Computer Science and Engineering
Speaker Name
Jitendra Kumar ( Rollno : 520cs1001)
Speaker Type
Student
Venue
Room No: CS323, 2nd Floor, Dept. of CSE, NIT Rourkela
Date & Time
13 Dec 2022 11.30AM
Contact
Prof. Bibhudatta Sahoo
Abstract

Recommender Systems have gained popularity in recent years due to their ability to quickly expedite users' selection processes. Traditional recommender systems mainly focus on providing recommendations to a user. It is not suitable for recommending an item to groups of users. A group recommendation system (GRS) address this issues of recommendation. GRS is popular in a few domains such as parties, tourism, movies, etc. The objective of a group recommender system (GRS) is to provide appropriate recommendations for each group member. A few research is reported in the GRS domain that satisfy each user requirement in a group. The state of art technique cannot adequately address the issue of group satisfaction. Therefore, the task of GRS can be divided into three subtasks: the formation of the group, rating prediction of individual members in a group, and aggregating them.  We propose a novel technique for rating prediction of individual member in a group on an item considering user's characteristics such as age, gender, and occupation. Item rating is the linear combination of features. Subsequently, user-item rating is computed by combining user's average rating and item rating. We also suggest a novel aggregation function named Tendency-based Aggregation (TA), which is built on the group's user and item characteristics. In this work, the proposed method relies on linear and non-linear preferences. We compare different methods for preference aggregation and group choice prediction based on weighting individual preferences in linear preference. The weight computing is based on the node centrality score. Multiple centrality techniques are analysed for score calculation. An optimum value of a in non-linear preference is provided. We introduce two new modelling strategies (Hybrid1 and Hybrid2) to improve the member satisfaction and recommendation for groups. The experimental results show that the non-linear remapping of preferences yields better group predictions and recommendations. To improve the group recommendation and group satisfaction, we introduce member inclination and item usefulness in a group. The proposed method is discussed in two ways by using the aggregate prediction approach of GRS. In the first case, we compute user inclination and item usefulness considering the whole dataset. In the second case, we compute member inclination and item usefulness using only group information. We introduce a novel aggregation strategy based on popularity and likeness (PLAS). It combines the predicted member rating into a group score. We suggest a method to create a group using age, gender, and occupation genres. The different sizes of groups are formed using the category of user genres. This study proposes two novel approaches to predict the group member preference. In the first approach, predicted rating is the linear combination of the best similar user. The new similarity method is introduced to find the best similar user of a grouped. In the second approach, we propose a novel hybrid personalized tendency approach (HPTA). HPTA means entropy and clarity are combined with a personalized tendency approach (PTA). Later, a novel aggregation linear neural modelling (LNM) is introduced. It combines individual member ratings to get final group preferences. LNM says that group preference is the linear combination of all the members' ratings. It combines individual member ratings to get final group preferences. Experiment is conducted on MovieLense-1M (ML-1M), MovieLense-100k (ML-100k), and NetFlix-1M (NF-1M) datasets. The proposed novel approach outperforms the existing state-of-art method in various measures like MAE, RMSE, SEG, and GSM.