Reinforcement learning based anaphora resolution method

Abstract

The invention discloses a reinforcement learning based anaphora resolution method, which comprises the following steps: data preprocessing: carrying out word segmentation, sentence segmentation, part-of-speech tagging, part-of-speech reduction, named entity identification, syntactic analysis and word vector conversion on text data to obtain candidate preceding words and analogy word related characteristics; constructing a neural network model: combining the characteristics of the word vectors and the relevant characteristics which can learn the fingering pairs and the relevant semantic information, better sorting and scoring the candidate preceding words and the fingering words, and finally obtaining an fingering chain; and using the trained model to carry out anaphora resolution, inputting text data, and outputting a resolution chain. According to the method, deep learning training is carried out by adopting a reward measurement mechanism for overcoming the defects of a heuristic lossfunction, the model effect is improved, hyper-parameter setting is automatically carried out for different language data sets, the necessity of manual setting is avoided, the practicability of the model is improved, and the application range is expanded.

Description

technical field [0001] The present invention relates to the field of natural language processing, more specifically, a method for anaphora resolution based on reinforcement learning. Background technique [0002] Reference is a ubiquitous expression in natural language. In order to avoid repetition, people are accustomed to using pronouns, titles, and abbreviations to refer to the aforementioned entities, which makes the language concise and coherent. However, a large number of references increases the difficulty of natural language processing, and reference resolution is the task of identifying different expressions of the same entity in the text. It has an extremely important basic support role for natural language processing applications such as information extraction, automatic summarization, automatic question answering, machine translation, and machine reading comprehension. [0003] There are mainly the following methods for referencing resolution: [0004] Resoluti...

AI Technical Summary

Problems solved by technology

Common reference resolution models include reference pair model, reference ranking model, entity expression model, etc. Usually, the reference pair model only extracts information from two independent words to determine whether the two words have a reference relationship. Far from enough, especially when the candidate antecedent entity expression lacks effective information description, the effect is even worse, so only using the features of entity reference pairs often eliminates the general effect

On the other hand, in the process of digesting model training, most models are trained using heuristic loss functions. For different languages ​​and different target data sets, it is often necessary to manually adjust the hyperparameters of the loss function, and the model generalization ability is not strong.

Images