RECOMMENDATION SYSTEMS : IMPROVING PERFORMANCE THROUGH HYBRID TECHNIQUES

This thesis presents recommender techniques, their strength, weaknesses, and the effectiveness of making up for the drawbacks of a recommender technique by exploiting the strength of one or more other techniques in a combination approach called hybridization. This is done to improve personalized recommendations. I demonstrate the hybrid technique with LighFM recommender algorithm as a scenario of items metadata complementing collaborative filtering to level out its weaknesses in a cold start scenario where collaborative interaction data is scanty. The result reveals that the hybrid model outperforms pure collaborative filtering where collaborative data is insufficient and at least as well as the pure collaborative filtering where collaborative data is sufficient