Method for personalizing chat bots
OwnerRichard Scot Wallaceinventor
Life of this patent
Estimating the term…
Abstract
A system and method for automatically building chat bot content from scripts of conversations. A script is parsed into query-response pairs, so that a computerized device may be programmed to give the same responses when given similar queries. Each query is converted into canonical form and then matched against patterns, as in prior art chat bots. The response is then linked to the matched pattern. A new chat bot is constructed out of the canonical form conversions and the set of pattern-response links. The chat bot is further improved by adding interactive human conversations, provided that those conversations are rated highly. The resulting chat bot can run on a computer server that responds to public queries, and the server can also display context-sensitive advertisements.
Claims16
A computerized method for personalizing a chat bot, comprising the steps:
- loading a query sentence and a response sentence;
- applying a set of reduction rules to said query sentence to get a canonicalized query sentence;
- matching said canonicalized query sentence against a set of patterns to get a matching pattern;
- combining said matching pattern with said response sentence into a pair;
- and adding said pair into said chat bot;
- wherein said query sentence for said chat bot is from a first user, and said response sentence is from a second user who is simulating a particular personality.
A computerized method for personalizing a chat bot, comprising the steps in sequence:
- loading a query sentence and a response sentence from a transcript;
- applying a set of reduction rules to said query sentence to get a canonicalized query sentence;
- matching said canonicalized query sentence against a set of patterns to get a matching pattern;
- combining said matching pattern with said response sentence into a pair;
- adding said pair into said chat bot;
- representing said set of reduction rules as a reduction tree with hashed branches for efficient depth-first searching;
- and
- representing said a set of patterns as a pattern tree with hashed branches for efficient depth-first searching.
A computerized method for creating chat bot content, comprising the steps:
- getting a reduction matcher and a pattern matcher from another AIML chat bot;
- parsing a conversation script into query-response pairs with a query sentence and a response sentence;
- for each of the query-response pairs, using said reduction matcher to canonicalize said query sentence, and using said pattern matcher to find a pattern that matches the canonicalized query response;
- building an AIML list of the patterns coupled with the response sentences;
- and creating a new chat bot by combining said AIML list with said reduction matcher.
Dependencies are inferred from the claim text.