Psychological Counseling Chatbot Based On GPT-2

Compared with real psychologists,psychotherapy chatbots have certain advantages,such as low cost,avoiding psychological barriers for psychological patients,not being limited by time and place,and not limited by the number of psychologists.Existing psychotherapy chatbots have been proved to significantly increase the chances of patients receiving therapy,and have certain psychotherapy effects.However,some unsolved technical requirements limit the use of psychotherapy chatbots,such as: how to avoid meaningless or vague chatty answers,how to avoid provocative language,how to integrate psychotherapy theory into the dialogue model,how to ensure the therapeutic effectiveness of psychological counseling chatbots,etc.The solution to these technical requirements is based on the collection of standard,effective,authentic,and rich psychological counseling corpus.Two basic and unaddressed technical needs are to avoid meaningless chatty answers and inconsiderate language.On the one hand,when users chat for psychological consultation,chatty answers tend to make patients with mental illness feel neglected,thus bringing greater psychological pressure.On the other hand,mentally ill patients are more sensitive.A sharp answer may have serious consequences.Therefore,compared with daily chatbots,the answers of psychotherapy chatbots need to be more targeted and considerate.The elimination of meaningless chatty answers and the definition and calculation of the thoughtfulness of the response of the psychotherapy chatbot can ensure the psychological safety of users,help to optimize the response strategy of the psychotherapy chatbots,and improve the user experience.The work of this dissertation mainly includes the following three aspects:(1)A large-scale data set of conversational data for psychological counseling is constructed,and a dialogue category judgment model is designed.The corpus related to psychological counseling is selected from a large number of self-collected corpora,and the corpus problem required by the psychotherapy chatbot is initially solved.(2)A response model based on GPT-2 pre-training model is constructed using the psychological counseling conversation corpus to realize the psychological counseling chatbot Medivh,and meaningless chatty answers are filtered out by maximum mutual information.(3)We define the evaluation standard of psychological counseling chatbots: response consideration.We construct the response consideration evaluation model(RCEM)for calculating the response consideration of psychological counseling chatbots,and several groups of comparative experiments are carried out to prove that the response consideration evaluation model can accurately calculate the response consideration of chatbots.