Personalized Recommender System Using Topic Modelling Approach

A personalized recommender system provides recommendations to users based on their preferences, behaviors, and past interactions with a platform. Data sparsity, cold start, bias, privacy concerns, lack of transparency and overfitting are the challenges in this system. There are several types of recommendation systems in the market. Some of the most commonly used systems are content based, collaborative based filtering, hybrid, knowledge and demographic based systems. Many recommendation systems rely on popularity-based recommendations, where popular items are recommended more often than less popular ones. This can lead to over-representation of popular items and limit the diversity of recommendations and at the same time may lead to overfitting as it’s tailored too closely to user’s behavior. Personality aids recommendation systems in comprehending user preferences and creating personalized suggestions that complement each user’s specific traits, increasing relevance and user contentment. It can bring novelty, fill in information gaps, and boost user engagement, creating a more satisfying and individualized experience. By employing effective techniques to manage a large amount of data and carry out computations fast, a personalized recommender system may be made scalable. To manage several users simultaneously, the task might be divided among several computers or servers. To save effort and enhance speed, it may also store and reuse previously computed suggestions. The system is hence capable of managing more users and data without slowing down or crashing. The proposed system used Topic modelling approach and MBTI (Myers-Briggs Type Indicator) personality classes to effectively suggest the user a restaurant that is suited for their personality and this is done by analyzing eWoM (Electronic Word-of-Mouth) of users already posted on the social platform. Extreme gradient boosting and similarity measures like Cosine similarity are employed here to match the user posts with various topics and ultimately to a personality class from the 16 available labels. This proposed system focuses on restaurant recommendation based on the grounds that each restaurant has a personality associated with it and people with similar personalities have similar tastes. It exhibits a precision of 81.8% hence vouching for the efficacy of the system and giving insight for further innovations and improvements down the lane.