Question answer system for online feedable new born Chatbot

Designing a new born Chatbot and feeding it from the web with specific areas of interest is a new research field. Few researchers have investigated empty database Chatbots and populating it from web pages or plain text corpora. Extracting data from web pages needs considerable processing before the response sentences are ready for the Chatbot. Feature extraction is also needed in order to filter and quantify the extracted plain text. In addition, ranking and classification are also required. This paper presents a new method that employs multiple feature extraction methods to quantify text responses for a new born (uneducated) Chatbot. Multiple measurement metrics are examined simultaneously in order to find the nearest match to a query. The nearest matches with the highest score have been obtained by re-ranking the scores of extracted features for text responses. The results show that the highest scored sentences have subjectively a good match to the query. The evaluation results indicate that the performance of the system increases significantly by using cosine similarity to find lexical match between the query and the response sentence rather than Jaccard's coefficient.