Aspect category-based interpretability recommendation method and system fusing external data

Abstract

The invention discloses an aspect category-based interpretability recommendation method and system fusing external data. The method comprises the steps of S1, training an aspect category classifier and an aspect category sentiment polarity classifier by utilizing external standard annotation data; s2, utilizing the aspect category classifier and the emotion polarity classifier to classify the comment data to obtain an aspect category vector [a1, a2,..., an] of each comment and an emotion vector [p1, p2,..., pn] corresponding to each aspect, wherein n is the number of aspects; and S3, fusing the aspect category vector and the emotion vector to obtain a prediction score and a recommendation reason of the to-be-recommended commodity. According to the method, external data is introduced, the accuracy of aspect and emotion polarity judgment is improved, comment modeling is more accurate and more standardized, and higher interpretability is achieved; the modeling of the supervision commentsis assisted by utilizing the information of the external standard annotation data, so that the cost of additionally annotating the comment information is reduced.

Description

technical field [0001] The present invention relates to the technical field of data processing, in particular to an explainable recommendation method and system based on aspect categories that integrate external data. Background technique [0002] The current mainstream recommendation systems are divided into collaborative filtering methods, content-based recommendation methods and hybrid models combining the two. Among them, collaborative filtering is the current mainstream practice, which can be divided into neighborhood-based recommendation methods and model-based recommendation methods. The neighborhood-based recommendation method includes collaborative filtering of users and products, which mainly uses similar users and products in the existing purchase history to make recommendations. Model-based recommendation methods are based on neural networks, matrix decomposition, ranking models, etc., using the user's historical behavior records to model, get the representation...

AI Technical Summary

Problems solved by technology

[0008] In the existing technical solutions, the recommendation method that does not consider reviews naturally lacks the user's subjective point of view, and even lacks the user's emotional information

In addition, for the current recommendation method combined with user comments: First, the handling of comments is rough and unclear, there is no external standard information to supervise, and additional comment labeling requires a lot of manpower and material resources to support, so from The user preferences and product attributes extracted from the reviews are abstract and rough, and the rich information in the reviews has not been fully utilized; second, the interpretability of the existing recommendation methods is relatively rough. Explanation is still limited to what most people like, what similar users like, or similar items that have appeared many times in user historical behavior records. Interpretability is equally important

Images