Project Summary

Jarvis is an assistant chatbot AI that will help players by answering their queries about various minecraft items via in-game chat. The goal of our AI is to achieve on-time input information resolving/analysis and return useful information as a response. The scope of knowledge for our Al will be tentatively restricted to “item crafting.” It will take in questions such as “where can I find this item” or “how do I craft this item” and use NLP to deduce what information the user wants, which will be outputted in chat.

AI/ML algorithms used

NLP (Natural Language Processing)

Evaluation plan

Quantitative

The baseline we will be comparing Jarvis to would be a simple search system that looks for a specific keyword in the user’s input and returns a general response based on the keyword it recognizes. We hope to improve the accuracy, input analysis and search relevance of Jarvis over this baseline. It would be nice if Jarvis is able to match the speed of our baseline, but we are not expecting it to do so if we want to improve our metrics.

Qualitative

For sanity cases, we will first verify that Jarvis can at least recognize certain keywords pertinent to items. From there, the next milestone will be to have Jarvis recognize a single basic question. The next milestone after that would be for Jarvis to recognize the same question worded in different ways. In the best-case scenario, we aim to have Jarvis be able to correctly interpret the player’s question with a 90% success rate or higher. This would involve recognizing the item being asked about and the information the player wants regarding said item. (Another way to evaluate the AI is to rank the input text based on difficulty. The raw input could be either plain/straight forward or ambiguous/multi meaningful. How capable the AI is could be tested based on which level of input it could resolve correctly in a rate)

Appointment

We will meet with the professor on 1/20 at 4:45 PM.