The number of chatbots that can be found in the web increases everyday and, alongside these, the amount of resources provided by the chatbot's community, which can be used to build new chatbots.
In this thesis we target to create a platform, Just.Chat, that helps people either to create the knowledge base of a chatbot from scratch or to enrich a previously defined one, by using available chatbots' interactions.
We start by building the Chat corpus, obtained from different sets of interactions from existing chatbots (some manually crafted). Then, we develop Just.Chat, which provides three different filters to automatically process these interactions, corresponding each filter to a distinct problem that could rise from crudely incorporating the Chat corpus into a chatbot. The first filter discards interactions overlapping others previously scripted in the chatbot knowledge base; the second filter is responsible for identifying questions and, among these, personal questions that the user may want to customize according to the chatbot's character in hands; finally, the third filter deals with interactions containing terms or topics that the user does not want his/her chatbot to use. All the filtered interactions are put aside, and is given to the user the opportunity to review and incorporate them in the chatbot's knwoledge base.
All these filters are individually evaluated, as well as Just.Chat, which is used to enrich the knowledge base of Edgar, a butler that has Monserrate's palace as its field of expertise. With the addition of the new interactions, processed by the Just.Chat platform, Edgar is able to properly deal with more 24% of interactions, considering a corpus of interactions built from real users in Monserrate.