Open ahtsham58 opened 4 years ago
Hi,
Indeed the generate_responses.py
script loads real dialogues from the dataset and generates answers whenever the human recommender speaks, i.e. after each seeker's utterance. In your example, the system would only take "hello, i am looking for a cartoon movie" and then generate a response.
You are correct in saying that the model has many inconsistencies. Not being able to process genres asked by the user is one of them. This is expected, since in this model the recommendations are only based on the sentiment related to the movies previously mentioned in the dialogue. Therefore if the user asks for a genre but does not specify a movie, this system cannot generate a proper recommendation.
Thanks for elaboration.
Dear @RaymondLi0,
I am figuring out the predicted response generated by the system and this is what I understood..
System takes previous seeker and ground truth sentences as input and predict the next sentence accordingly. For example, if seeker say "hello, i am looking for a cartoon movie", ground truth: "fantastic ! have you seen ponyo yet ". So in this case, your system will take both utterances as input and predict the next sentence ... Am I correctly understood the recommendation approach?
If so, I observed many inconsistencies in the generated responses, e.g., in this situation, system generate: " have you seen thor : ragnarok ?" which not even close to cartoons instead its about Action, Adventure, Comedy, Fantasy, Sci-Fi according to IMDB genre classification.
Anyways please confirm the recommendation approach.. your cooperation would be appreciable and looking forward to your positive response.
thanks in advance!