Persons of Interest: A new Knowledge Base for Building "Creative" Bots

I have recently released a large knowledge-base (approx. 30k triples) of pop-culture facts and figures for the PROSECCO code-camp in Coimbra, with the intention of also using that knowledge for automated tweet generation and story-telling. This note describes the current state of the knowledge-base (which I call the NOC List, for Non-Official Characterizations List, in a nod to the McGuffin in the first Mission Impossible movie).

I have also written some easy-to-use Java classes for accessing and working with this knowledge in Java, which I will describe in this blog post. With this data and these tools, even novice (Java) programmers will be able to do interesting idea generation in the pop-culture realm. I illustrate below how I use this code to generate a variety of what-if tweets that form part of @MetaphorMagnet’s repertoire.

First off, the latest version of the knowledge-base and code can be accessed in our Githib repository here.

This repository contains a README directory that contains the slides on the NOC List that were used for the code-camp, and a PDF of my introductory book (Hand-Made By Machines).

This repository also contains a CODE directory that contains the Java code for accessing the knowledge-base(s) and for using the knowledge-bases to generate tweets for @MetaphorMagnet. More of this below.

This repository also contains a DATA directory with two sub-directories that encode the knowledge in two-different formats. The TSV Lists directory stores each knowledge-base as a text file (.txt) that is tab delimited. The first line contains the field names for the file, also tab delimited. The directory called Veale’s The NOC List contains the corresponding data files in an easier-to-edit/browse spreadsheet (.xlsx) format.

The slides for the code-camp (in the README directory) offer the best overall introduction to the knowledge-bases. A more detailed perspective is offered by the code itself. So let’s look more closely at that.

The class is designed so that a single instance contains all the information in a single spreadsheet/TSV file. So a different instance is created for each of the data files in the knowledge-base. This is an extract from the class, which uses the data to generate interesting things:

NOC = new KnowledgeBaseModule(kDir + “Veale’s The NOC List.txt”, 0);

CATS = new KnowledgeBaseModule(kDir + “Veale’s Category Hierarchy.txt”, 0);

CLOTHES = new KnowledgeBaseModule(kDir + “Veale’s clothing line.txt”, 1);

CREATIONS = new KnowledgeBaseModule(kDir + “Veale’s creations.txt”, 0);

DOMAINS = new KnowledgeBaseModule(kDir + “Veale’s domains.txt”, 0);

WORLDS = new KnowledgeBaseModule(kDir + “Veale’s fictional worlds.txt”, 0);

VEHICLES = new KnowledgeBaseModule(kDir + “Veale’s vehicle fleet.txt”, 1);

WEAPONS = new KnowledgeBaseModule(kDir + “Veale’s weapon arsenal.txt”, 1);

PLACES = new KnowledgeBaseModule(kDir + “Veale’s place elements.txt”, 0);

SUPERLATIVES = new KnowledgeBaseModule(kDir + “superlatives.txt”, 0);

COMPARATIVES = new KnowledgeBaseModule(kDir + “comparatives.txt”, 0);

ANTONYMS = new KnowledgeBaseModule(kDir + “antonyms.txt”, 0);

PAST_PERFECTS = new KnowledgeBaseModule(kDir + “past perfects.txt”, 0);

Choose kDir to point to the TSV Lists subdirectory of the DATA directory and the above modules should load cleanly.

The last two data modules are a useful source of glue data for putting everything together in an NLP output.ANTONYMS maps properties (such as strong) onto their opposites (such as weak), while PAST_PERFECTS maps the present continuous form of action verbs (such as shooting) onto their past perfect forms (e,g. shot).

Have a look inside to see how these knowledge modules are used to generate a variety of different tweets, often anchored in an opposition of properties (provided by ANTONYMS).

For instance, the method generateDreamConflicts(…) generates tweets about dreams in which pop-culture figures appear. Here is an example of a two-parter:

Last night I dreamt I was cleaning floors with #GroundskeeperWillieMacDougal when we were run over with ruthless ambition by #HillaryClinton

I guess #HillaryClinton and #GroundskeeperWillieMacDougal represent warring parts of my personality: the capable vs. incompetent sides.

[The ANTONYMS module provides the crux of the dream here, the contrast between capable and incompetent]

The method generateNietzscheanTweets(…) generates what-if tweets that riff on Nietzsche’s famous line “What doesn’t kill you makes you stronger“, as in:

If what doesn’t kill you makes you stronger, shouldn’t being overwhelmed with ruthless ambition by #HillaryClinton make you more driven?

If what doesn’t kill you makes you stronger, shouldn’t being knocked out with an Oscar statuette by #DanielDayLewis make you more talented?

The method makeOthersLookGood(…) generates a relativistic comparison, where the speaker believes he/she resembles X but others say they resemble Y, where X and Y differ in some antonymous respect. Here’s an example:

I see myself as capable, but my boss says that I make even someone as incompetent as #EdWood look like #HillaryClinton.

The method generateShakespeareanTweets(…) generates what-if tweets that riff on the Shakespearean idea that “Clothes maketh the man”. Here is an example:

If clothes maketh the woman, would wearing #HillaryClinton’s pant suit make you more ambitious? Or more grasping?

The method walkMileInShoes(…) generates what-if tweets that riff on the old phrase “you shouldn’t judge someone until you’ve walked a mile in their shoes”. Here’s an example or two:

Nobody’s perfect! My grandma says to never judge a grasping first lady like #HillaryClinton until you have walked a mile in her pant suit.

Nobody’s perfect! My grandpa says to never judge a shallow diarist like #CarrieBradshaw until you have walked a mile in her Manolo Blahniks.

Nobody’s perfect! My mom says to never judge a neurotic actor like #JackNicholson until you have driven a mile in his Jokermobile.

The class PersonOfInterest also contains methods to generate different kinds of XYZ metaphor (i.e. of the form X is the Y of Z). Here are some examples:

What if #TheEmpireStrikesBack were real? #HillaryClinton could be its #PrincessLeiaOrgana: driven yet bossy, and controversial too

What if #TheNewTestament were about #AmericanPolitics? #MonicaLewinsky could be its #Lucifer: seductive yet power-hungry, and ruined too.

If #MonicaLewinsky is #Lucifer in a stained blue dress, who in #TheNewTestament is #HillaryClinton most like?

When it comes to #TheDemocrats, is #HillaryClinton just #UriahHeap in a pant suit? She is ambitious yet grasping, and controversial too

Well, you get the idea. I hope this data and this code make it easy enough for you (or most likely your students, at undergraduate or postgraduate-level) to generate some interesting pop-cultural outputs for use in stories, jokes, tweets, etc. Let me know if you encounter any problems with the data or the code. Certainly feel free to add to the data (or correct mistakes you may find), and let me know if you generate anything interesting with it. Comments below are also welcome.

Scholarly Lite is a free theme, contributed to the Drupal Community by More than Themes.