The bot can be launched using different algorithms to generate answers and specifying how the bot will be used. There are three algorithms that can be used to specify the answer: --answergenerator, --conceptextractor, --seq2seq. There are three ways to use the bot: --onlyquestion, --onlyanswer, --bothQA.

• --answergenerator: doesn't make any use of learning algorithms;
• --conceptextractor: uses two neural networks, one to classify the relation of the question and another one to extract concepts;
• --seq2seq: uses Seq2Seq algorithm (end-to-end approach).
• --onlyquestion: the user can only ask questions (querying);
• --onlyanswer: the user can only answer to the questions of the bot (enriching);
• --bothQA: both querying and enriching.

To run the bot it's necessary to specify a JSON file that describes the network, the algorithm to answer to the questions and the way the bot will be used. For example, to run the bot that does both querying and enriching with concept extrator algorithm, it's necessary to type:

python3 bot.py ../models/hparams.json --conceptextractor --bothQA

The username of the bot on Telegram is @cip_nlp_chatbot.

To train the networks it's necessary to specify the structure of them with a JSON file, for example:

python3 train.py ../models/hparams.json

Note that you need to have both Keras and PyTorch. Moreover, to train only some of the networks you need to modify the flags TRAIN_* from the code. The models will be saved in the folder models.

To download the knowledge base simply type from the terminal:

python3 download_kb.py

The knowledge base will be saved on a JSON file in the folder resources.

To generate a vocabulary from the knowledge base simply type from the terminal:

python3 vocabulary_from_kb.py N

where N is the minimum number of times a word has to appear in the knowledge base to be included in the vocabulary. The vocabulary will be saved in the folder resources.

To generate a unique question pattern file from multiple files type from the terminal:

python3 patterns.py

Note that before running this script you have to remove bad files and after the execution of the script you have to double check the file created in the folder resources by hand.

To debug the bot a simple script has been developed to search for words or to select a random tuple randomly. Run the script by typing:

python3 query_kb.py

Once the program has been loaded the knowledge base type search=SENTENCE to search for a sentence, random to randomly select a tuple or quit to exit from the program.

• Keras (built on top of Tensorflow)
• PyTorch
• Telepot
• Gensim