Water conservancy portal information recommendation method based on multi-layer attention mechanism and fused with map

Abstract

The invention provides a water conservancy portal information recommendation method based on a multi-layer attention mechanism and fused with a map. For the characteristics of collected water conservancy information data, Doc2vec and clustering are utilized to enrich a feature set; when a recommendation model is constructed, firstly, a feature-level attention mechanism is formed by means of softtention to fuse all features, meanwhile, a water conservancy information graph is constructed, and after potential interests of a user are mined, a final representation vector of each piece of water conservancy information is formed; then, a behavior-level attention mechanism is formed by using a self-attention with a position code to generate a representation vector of a user behavior; a representation vector of the interest of the user is generated by means of soft-attention; and finally, the probability that the user clicks the water conservancy information is calculated by using a multi-layer perceptron to generate a final TOP-N recommendation list. According to the method, the problem that personnel engaged in the water conservancy industry cannot read interested water conservancy information in a one-stop mode can be solved, and recommendation can be more accurate and higher in interpretability by means of a multi-layer attention mechanism and atlas.

Description

technical field [0001] The invention relates to a water conservancy portal information recommendation method, in particular to a water conservancy portal information recommendation method based on a multi-layer attention mechanism and fused maps. Background technique [0002] The water conservancy portal is a platform for the water conservancy department to integrate and release water conservancy information and provide information and services to the public. It is oriented to the water conservancy industry and society, and is used to report various water affairs trends, convey and implement the spirit of relevant documents and major conferences of the Ministry of Water Resources. After the continuous development of water conservancy informatization, there are now many water conservancy websites, and each river basin, province, and even district and county level has its corresponding water conservancy website. However, these systems are often independent of each other, and ...

AI Technical Summary

Problems solved by technology

Wang et al. used knowledge graphs to enrich the information in news entities and strengthen the connection between entities. However, the features used are too single, which will cause information loss and even produce "Simpson's paradox" to damage the performance of the recommendation system; Zhou et al. Humans proposed a deep interest model to predict the click-through rate of products. They used the attention mechanism to weight and aggregate each clicked product to distinguish the importance. However, the click sequence is usually time-related, and the model does not take into account the relationship between each click behavior. Dependency; Wu et al. proposed a recommendation method with attention and multi-view learning, using a three-layer attention mechanism to encode text, news and users respectively, but for users, whether the news that has been clicked can be corrected The current news to be recommended has a certain influence, and there are insufficient considerations in this regard

Images