Text-based Collaborative Filtering For Cold-Start Soothing And Recommendation Enrichment

Created: 2022-03-29 11:43
#paper

Main idea

The main idea behind this paper is to build a unified vector space where both users and items are mapped. In this way similar items and users are mapped close together (if they use similar words), so that a classical neighborhood-based Collaborative filtering scheme can be used to recommend items. The model is called Conceptual Skip-Gram (CSG).
.

In deep

Objectives:

• words co-occurring in the reviews should be close to each other;
• concepts (users and items) should be close to their words;

In a Cold-start setting, any textual content can be used to create a profile by simply optimizing the objective functions for each word from the data.

A review (in our case a #post) can be extracted as the sum of the representation of its words.

Another feature of this system is the ability to justify each recommendation by extracting the nearest words of an item.

References

1. Paper
2. Word2Vec

Code

Tags

#word2vec #csg #embedding #cold_start