Service discovery method and system based on long short-term memory network under attention mechanism

Abstract

The invention provides a service discovery method and system based on a long short-term memory network under an attention mechanism, and the method comprises the steps: taking a word vector of a sentence pair as the input of a BiLSTM layer to construct a feature vector, constructing the internal relation of the sentence pair through employing the attention mechanism, and carrying out the context feature vector extraction of the sentence pair information after the operation of the attention mechanism; obtaining a vector with a fixed length through pooling operation and serving as input of an output layer, and predicting the matching level of sentence pairs; optimizing the model; carrying out similarity matching on all given query requests and the test set service; according to the predicted probability, sorting the matched services in an inverted order, wherein the first N services with the highest score are target services to be retrieved, and in combination with the advantages of a BiLSTM language model and an attention mechanism, feature vectors containing rich context semantic information are generated based on combination of forward semantics and reverse semantics of the BiLSTM model; and supporting service similarity calculation effectively by using an attention mechanism, and finding the most accurate target service.

Description

technical field [0001] The invention belongs to the field of computer science and technology, and specifically relates to a service discovery method and system based on a long-short-term memory network under an attention mechanism. Background technique [0002] The main task of service discovery is to match the functional description information between query requests and candidate services. This information is usually represented in the form of natural language text. Traditional neural network modeling, such as DNN modeling, is essentially a bag-of-words model that cannot distinguish word order, but because word order actually has a great impact on sentence meaning, it cannot correctly reflect the semantics of sentences. For the CNN neural network, only local n-gram information is extracted, and the word order of the sentence, especially the long-distance word order, is indistinguishable. This has limitations for feature vector matching of long sentences. In addition, alt...

AI Technical Summary

Problems solved by technology

Traditional neural network modeling, such as DNN modeling, is essentially a bag-of-words model that cannot distinguish word order, but because word order actually has a great impact on sentence meaning, it cannot correctly reflect the semantics of sentences

For the CNN neural network, only local n-gram information is extracted, and the word order of the sentence, especially the long-distance word order, is indistinguishable, which is limited for feature vector matching of long sentences.

In addition, although the commonly used similarity calculation method based on space vector is simple and easy to implement, it lacks comprehensive comparison and matching from multiple perspectives, which affects the accurate retrieval of services

Images