WatsonDetective

Inspiration

This project was inspired by a simple fascination with the Watson API, in all of its complexity. We found that the capabilities of the Watson API were wide in scope, and incredibly responsive. We also shared a common cybersecurity interest, and endeavored to create a Watson conversation bot that could intelligently handle some aspects of a cyber investigation.

What it does

The Watson API is mostly leveraged using a LogParser Python object, which generates a running commentary of Watson's insights on the subject's emotions. Watson has a conversation workspace created with basic dialog focused on ascertaining if a user was on his (or her) computer during a given time-frame.

If the user responds negatively, Watson attempts to backtrack and apply a different emotional setting. Continued hostility (to anomalous levels) results in an email being generated for the user that is managing Watson so a separate conversation can be struck up to acquire the information.

How we built it

Watson is currently a pure-Python3 program that leverages the Watson SDK, as well as matplotlib's pyplot.

Challenges we ran into

Parsing Watson is hard! He provides a whole slew of information in JSON format, and most of the information we were looking for was deep within the JSON structure. Matplotlib also proved to be a difficult beast to handle.

Accomplishments that we're proud of

This code, specifically.

` def tone_report(self, text, tree=None): if not tree and tree != []: self.muse("acquiring tone") resp = self.tone_client.tone(text=text)

        tree = resp['document_tone']
        self.muse(tree)

        # Check for warnings and print them if they exist.
        if 'warnings' in resp and resp['warnings']:
            self.muse(resp['warnings'])

    if 'tone_id' in tree and 'score' in tree:
        self.muse("yielding")
        yield tree['tone_id'], tree['score']
    for branch in tree:
        if isinstance(tree[branch], list):
            for twig in tree[branch]:
                for name, percentage in self.tone_report(None, twig):
                    self.muse("recursive call")
                    yield name, percentage
        else:
            self.muse("not list")`

It allows us to recursively search through a Tone Analytics dictionary and generate tuples that directly connect each emotion to its score.