The ShopBot project is a proof-of-concept for tagging shoppers' queries with their objective.
Running ShopBot requires Python 3.5 with the following packages installed:
scikit-learn, pandas, klein and nltk with punkt tokenizer models.
Execute the following commands, if needed:
$ pip install scikit-learn pandas klein nltk
$ python
>>> import nltk
>>> nltk.download('punkt')
Alternatively, run python install.py.
To try ShopBot, clone this repository, cd into the main folder and run
python bot.py [options]
Options include:
| option | meaning | default |
|---|---|---|
--port |
web server port | 8080 |
--file |
tagged samples file to load | tagged.csv |
-K |
number of similar docs to consider | 3 |
The service exposes the following REST API operations
POST /v1/tags where the request content is the query to tag.
The query is expected to be a single sentence, and the service returns a single tag
with its probability in JSON format.
$ curl -s -d "I want a lightweight phone" "http://localhost:8080/v1/tags"
{"tagging": "weight", "probability": 0.692, "text": "lightweight phone"}$ curl -s -d "does it have enough RAM?" "http://localhost:8080/v1/tags"
{"tagging": "memory", "probability": 1.0, "text": "does it have enough RAM?"}$ curl -s -d "I want to take selfies!" "http://localhost:8080/v1/tags"
{"tagging": "camera", "probability": 1.0, "text": "I want to take selfies!" }PUT /v1/tags where the request content is a JSON object with text and tag
fields. The service will add the tagged sample to its documents set.
The operation returns 'OK'
$ curl -s -X PUT -d '{"text": "Java applets", "tag": "java"}' "http://localhost:8080/v1/tags"
OK
$ curl -s -d "I like java games" "http://localhost:8080/v1/tags"
{"tagging": "java", "probability": 0.542, "text": "java phone"}Read the full report for the implementation details, algorithms and background.