nlp-chatbot

Stack Overflow Assistant Telegram Chatbot, based on templates from Potapenko et al, HSE, Russia.

Accessing the bot:

Simply go to t.me/prannerta100_bot and start chatting!

The bot has been hosted on a Google Cloud VM, so you might face some server-side latency.

Here is a sample conversation with the bot:

Dependencies:

1. nltk
2. chatterbot corpus
3. sklearn

Description:

Here is how the bot works:

1. read the message, use a classifier to classify its intent as non-programming or programming
2. If non-programming, then invoke a pre-trained, off-the-shelf chatbot (here I used Ron Obvious from the chatterbot module)
3. If programming, then show the relevant Stack Overflow link that matches the question description the best

Training data:

• The first classifier (programming versus non-programming) was trained by mixing 2 datasets:

1. tagged_posts.tsv — StackOverflow posts, tagged with one programming language (positive samples).
2. dialogues.tsv — dialogue phrases from movie subtitles (negative samples).

• The second classifier (given a programming question, find the relevant Stack Overflow link) works by finding the closest StarSpace embedding.

Relevant files:

• The most relevant file here is nlp-chatbot/nlp-chatbot-main/dialogue_manager.py. The entered question is transformed into a TF-IDF feature vector. A model is trained to identify the most relevant question and tag.

Send feedback to prannerta100@gmail.com.