Research On Text Sequence Annotation Based On Multi-task Collaborative Learnin

Text sequence tagging is an important research direction in the field of natural language processing.Its main purpose is to label every word or phrase in the text so as to understand the meaning of the text more deeply.It provides a basis for the construction of tasks such as machine translation,knowledge mapping and question answering system.In order to solve the problems of wasting resources,ignoring inter task relationships,and limited generality in traditional single task learning models,the research on text sequence annotation based on multi task has become the main direction of current research.At present,there are two main difficulties in the research of multi task text sequence annotation: 1.The problem of feature representation in multi task learning;2.The problem of inter task relationship collaboration in multi task learning.This article focuses on multi task learning as the research focus,mainly solving the two difficult problems currently faced by multi task sequence annotation.In the representation learning stage of multitasking,our goal is to fully explore and utilize the implicit relationship information between tasks,and structure the representation to enhance the final feature representation.Afterwards,by exploring the collaborative relationship between tasks and improving the collaboration mechanism between tasks,the overall effectiveness of the model is ultimately improved.In response to the above two difficult issues,this article takes the representation of multi task text features and the acquisition of relationships between multi task tasks as the starting point.The main research content is as follows:1.Research on text feature representation.Utilize multitasking learning to mine and utilize the relationship information between texts,and solve the problem of insufficient text feature information acquisition by utilizing the latent representation information between different texts.However,current research lacks sufficient utilization of latent semantic relationship information in text sequences,neglecting the relationship characteristics between sequences.Therefore,in order to improve the model’s ability to understand text and fully capture the hidden relationship information between texts,this paper proposes a hidden relationship enhancement model based on relational attention.This model uses structural alignment to obtain and maintain the Semantic information structure in the text sequence,and models the semantic relationship between texts,so as to improve the model’s ability to capture the relationship information between text words.In the end,the model achieved good results on multiple public datasets.2.Research on obtaining inter task relationships.In multitasking learning,there is a certain relationship between different tasks,such as part of speech tagging and named entity recognition.Part of speech information can better help identify the boundaries of entities,and in turn,entity tagging information can also better promote the accuracy of part of speech recognition.Currently,the relationship acquisition between multiple tasks is mainly achieved through sharing between tasks,with hard sharing being the most commonly used method.However,current sharing mechanisms all suffer from insufficient acquisition of inter task relationships and a lack of filtering of inter task relationship information.Therefore,this article proposes a multi task sequence annotation model for inter task collaboration.This model captures the relationship information between multiple tasks by modeling the abstract relationships between tasks.At the same time,in order to reduce mutual suppression between tasks and the clutter of shared information,adversarial training is adopted to retain useful information for the model,as well as filtering shared information.Finally,the feature representation of the task and the relationship information between tasks are fused to improve the extraction ability of the model.In summary,the multi task text sequence annotation model based on text representation and task relationships proposed in this article has achieved certain results in sequence annotation tasks,and sufficient experiments have been conducted on multiple public datasets to verify the effectiveness of the model.