MathChatBot

This is a project that I chose as final homework for the NLP laboratory at the Security and Applied Logic Master's program at the University of Bucharest.

Description

The project implements a (rudimentary) chatbot which (tries to) hold a conversation with you on mathematical topics. It will give you definitions of mathematical objects, ask you to study or to practice doing some exercises... y'know, nagging professor stuff.

How to run it

Open a terminal and run Python 3 on the file, specifying the output text file for logging.

Example:

python3 mcb.py log.txt

Then the conversation will take place in the terminal and the transcript will be available in log.txt.

If the program is not run with the output file argument, you will have the choice to add it or continue without it or exit.

How does it work?

The program uses the WordNet database for all the words and definitions, as well as the Stanford POS Tagger for parsing phrases.

The bot greets you and asks you what do you want to learn (or talk) about.

It has a database consisting of a list of synsets for mathematical terms.
It also has a database of specific verbs (e.g. learn, study, exercise, practice, define).
It parses the line you enter, looking especially for nouns and verbs (rather, noun phrases and verbal phrases), as tagged by the Stanford POS Tagger.
For each noun it found, it will make a list of similarity scores of it with respect to the nouns in the database, which will be sorted, to choose the one that is most relevant.
For each verb it found, it will look it up in the database.
Corresponding to the VP and NP, it will offer predefined answers.
If you are trying to define a polysemantic noun, it will offer all the definitions it has, until you are happy with them.

Example conversation

=== mcb.py ===
=== Fri Dec 28 11:29:15 2018 ===
Hi, how can I help you today?
Here are the topics I know something about:
group
limit
sum
product
integral
derivative
differential
exponential
logarithm
fraction
calculus
analysis
I can help you { study, learn, practice, read, exercise, define } some of these.
You can always end the conversation saying goodbye.
--> define group
I can give you the definition of group.
I know 2 definitions of group.
Here is one of them:
> Definition: any number of entities (members) considered as a unit
Is this what you were after?
--> yes
Great then!
What else do you want to talk about?
You can always end the conversation saying goodbye.
--> practice derivatives
Practice makes perfect, you know?
First, read the theory, then try solving some exercises in the textbook. See how that works for you.
What else do you want to talk about?
You can always end the conversation saying goodbye.
--> learn calculus
So, you want to learn about calculus or something related.
You can get together with some colleagues and try studying through discussions and debates if that works for you.
What else do you want to talk about?
You can always end the conversation saying goodbye.
--> I want to study products
So, you want to learn about intersection or something related.
Reading your course notes is always a good start.
What else do you want to talk about?
You can always end the conversation saying goodbye.
--> bye

Further features (F) and ideas (I)

(I) The bot will provide clickable links by querying DuckDuckGo !w for what you want to learn about.
(I) Add more terms, verbs and senses.

Known issues

Sometimes, the name of a noun does not correspond to what you entered. For example, for sum, it will also include union, so although sum is polysemantic, if it chooses union, it will not give you more senses. This can be fixed by disambiguating and/or checking more carefully for names and lemmas.

Tools used

NLTK, Stanford POS Tagger, WordNet in Python 3;
Emacs, Vim, st, GitLab.

Other credits

I learned to use the tools with the help of the NLP Lab (in Romanian) taught by Mrs. Irina Ciocan.
Inspired by ELIZA, see Common Lisp code here.

adimanea/mathchatbot