Research On Text Keyword Extraction Based On Eye-tracking Data

Text keyword extraction technology is to extract the words or phrases in the text that can best express its topic information.It is of great significance in many natural language processing tasks such as text classification and automatic summarization.The eye movement data records the reader's eye movement when reading the text.According to the difference in the importance of the text content,the reader's eye movement will also change accordingly,so the value of the word can be judged by the reader's eye movement characteristics,So that the keyword extraction model pays more attention to the key content of the sentence during training,thereby improving the performance of the model.The limitations of eye movement data collection technology make it difficult to apply eye movement features on a large scale,so some studies have proposed to use open source eye movement data sets for keyword extraction research and achieve certain results.The current research work still hasn't solved the three key problems in applying eye tracking data to keyword extraction tasks.First,the existing research only uses a single eye movement feature to measure the reader's attention to words,and does not consider combining multiple eye movement features.Secondly,the different sources of eye movement data and test data will lead to a considerable proportion of vocabulary missing the corresponding eye movement characteristics,but there is no research to propose effective solutions.Finally,the existing research lacks in-depth analysis of the effect of eye movement features in the keyword extraction task.Therefore,this article will conduct a more in-depth study on how to use eye movement data in keyword extraction tasks from three perspectives:multiple eye movement feature combinations,eye movement data expansion,and eye movement feature action effect analysis.In order to solve the problem of eye movement feature fusion,two methods of eye movement feature fusion are proposed.Through the experiment,it is found that the combination of multiple eye movement features is better than single eye movement features,and the method of using eye movement features as attribute features of words is better than that of attention mechanism.Secondly,in order to solve the problem of sparse eye movement features caused by the limitation of the size of eye movement data sets,this paper puts forward the corresponding solutions from two aspects: the merging of eye movement data sets and the expansion of eye movement data sets.In the expansion of eye movement data,this paper evaluates the word without eye movement feature from two aspects of word morphology and word meaning,and achieves certain results.Finally,this paper analyzes the influence of eye movement features from the frequency distribution of words,the numerical distribution of eye movement features and the distribution of eye movement features at the sentence level.The experimental results show that the distribution of eye movement features is an important factor that affects its influence in keyword extraction task;the more obvious the distribution level of eye movement features in numerical value,the better its effect;under the current coverage of eye movement feature words,the more important the role of eye movement feature is to indicate the non key content in the sentence,which in turn highlights the key part of the sentence ?In this paper,eye movement feature combination and eye movement feature expansion are used to solve the problem of single feature category and sparse feature when eye movement data is introduced into text keyword extraction task,which improves the performance of keyword extraction model.At the same time,this paper analyzes the reasons of eye movement characteristics,which has a certain reference value for the related research of eye movement data.