Search result diversification method based on generative adversarial network

Abstract

In the invention, a search result diversified training method based on the generative adversarial network is realized through a method in the field of artificial intelligence. The method comprises thefollowing steps: after giving out a query word, defining a corresponding candidate document set, and a sampler, a generator and a determiner unit which are sequentially arranged for a logic path; setting diversified scoring functions in the determiner and the generator, and carrying out training through a positive feedback process; introducing a generative adversarial network, and combining an explicit model and an implicit model through the generative adversarial network, wherein the final generator can generate better diversified document sequences through the means.

Description

technical field [0001] The invention relates to the field of artificial intelligence, in particular to a search result diversification method based on a generative confrontation network. Background technique [0002] Diversification of search results is an effective method to solve fuzzy queries raised by users. Most of the current mainstream diversification algorithms are supervised methods, which require high-quality data sets for training search result diversification models. The main goal of diversification of search results is to make the sorted list returned by the search engine cover all the subtopics of the user query as much as possible. At present, researchers have proposed a series of search result diversification algorithms. The main process of these algorithms is: when the user proposes a query word, according to the diversification scoring function, continuously select the best diversified document under the currently selected document sequence and add it to t...

AI Technical Summary

Problems solved by technology

In [2], the first 20 documents in the ideal sorting are used. Although the quality is high, the number is small, which may cause the model to underfit.

The scarcity of high-quality training data samples will lead to insufficient training or offset, affecting the final effect

At the same time, the existing models can basically be divided into the explicit model and the implicit model, without combining the two ideas, and without using all the information, if the explicit model and the implicit model can be combined in a certain way The combination of models may improve the diversification effect of search results

Images