persinammon / fantastic-books-in-clips

Chatbot implemented as expert system to recommend fiction to the casual reader, written as set of rules in LISP variant

Fantastic Books and Where To Find Them

An expert system is one of the first successful forms of AI and was popular in the 80's (yes, pre-Python and Java design patterns, but after the start of great ML and AI academic research). This expert system uses a rule engine written in late 1995 by an engineer at Sandia National Labs.

The rule engine itself is based on an implementation of the Rete algorithm, which optimizes on a simple looping through conditionals by implementing a trie of left hand side patterns to match, and marking nodes as they are fulfilled (not necessarily in sequential order). When a leaf node is reached, the corresponding rule is fired.

Dependencies

• Jess rule engine, Java-based, can integrate with JSR94 rule engine API
• CLIPS functional programming language
• Maven build system, install dependency by downloading jess.jar, running mvn install, referencing under a <dependency> tag in pom.xml

Functionality

• Asks series of questions based on characteristics of books in dataset
• Guaranteed recommendation when one remaining book that fits user written characteristics in dataset
• Uses a CLI and string-built questions

How to Run

This expert system is currently standalone, so this is the process to run the CLI recommendation system. A future improvement is to write a driver to run the code using Java then package into a .jar file.

1. Download jess.exe from https://www.jessrules.com/jess/download.shtml. (Fun fact: I got to interact with the Jess creator during this process).
2. Save book_recs.clp in the examples/jess folder.
3. Run jess.exe using bin/jess.
4. Run (batch "examples/jess/book_recs.clp").

Future Improvements

This program is not actively worked on at the moment, but forks and pull requests are certainly welcome. The following example extensions are not time consuming to implement.

• Move build to Maven and write Java driver.
• Add length of book as a characteristic.
• Make it so that if the first question is given a book the system doesn't know, it saves that book into the database.
• Standardize which characteristics go with which appeal factor.
• Output the reason why the book was chosen along with the recommendation.

Easy Contributions

• Add a book! Genres can be verified by looking at the Wikipedia page of the book.
• Update the characteristics of a book.