PERSONALIZED TRAINING RECOMMENDATION SYSTEM BASED ON COLLABORATIVE FILTERING

In this article we have covered many approaches of implementing personalized training recommendation system based on collaborative filtering. These techniques are consist of memory based methods, where we apply our statistical methods to the entire dataset to make predictions. We have considered such algorithms as cosine similarity and Pearson correlation. For cosine similarity we consider users data as vector of some collaborations in N dimensional space, where N is number of items. Then we calculate similarity of any two users as cosine of an angle between their vectors. This technique end up with good results, but anyway there is a problem, because of the matrix sparsity (empty interactions). Considering them as 0, impacts results even if we remove mean from each existing collaboration. Therefore, we also considered Pearson correlation which operates better with empty spaces in our data matrix. Here we try to find positive or negative trends between users and get correlation coefficient to predict rating.
At the end of article we have compared all techniques based on such approaches as measuring RMSE and MAE

1. Francesco Ricci and Lior Rokach and Bracha Shapira (2011), Introduction to Recommender Systems Handbook, 1-35

2. Ferrari Dacrema, Maurizio; Cremonesi, Paolo; Jannach, Dietmar (2019), Are We Really Making Much Progress? A Worrying Analysis of Recent Neural Recommendation Approaches, 101–109

3. He, Xiangnan; Liao, Lizi; Zhang, Hanwang; Nie, Liqiang; Hu, Xia; Chua, Tat-Seng (2017), Neural Collaborative Filtering,173–182

4. Fleder, Daniel; Hosanagar, Kartik (2009), Blockbuster Culture’s Next Rise or Fall: The Impact of Recommender Systems on Sales Diversity, 697–712.

5. Shi, Yue; Larson, Martha; Hanjalic, Alan (2014), Collaborative filtering beyond the user-item matrix: A survey of the state of the art and future challenges, 1–45