Method for matching queries with answer items in a knowledge base
The present invention includes an expert system in which a search index furnishes answers to incoming queries provided in natural language. A search index for a specific field contains components that facilitate selecting a best fitting stored answer to the incoming query. Furthermore, context of the incoming query (e.g. location of the user, a current web page or service being used/viewed by the user, the time, etc.) may be considered when selecting a best fitting answer.
1. A system for automatically matching a stored response to a user Natural Language Query (NLQ), said system comprising:
a data storage containing indexed stared responses to natural language queries, each indexed stored response being associated with an individual set of natural language words and indexed based on the individual set of natural language words;
a computing platform including processing circuitry adapted to:
receive digital data representing the user NLQ;
assign weight values to words in the user NLQ and to words in the individual sets of natural language words, based on a rate of occurrence of each of the words in at least one knowledgebase and rates of occurences of synonyms of each of the words in the at least one knowledgebase, such that a weight value assigned to a given word is inversely related to a rate of occurrence of the given word in the at least one knowledgebase and is further inversely related to rates of occurrences of synonyms of the given word in the at least one knowledgebase;
identify responses stored in said data storage having an associated set of natural language words including words appearing in the user NLQ;
identify responses stored in said data storage having an associated set of natural language words including synonyms of words appearing in the user NLQ;
calculate a query specific significance value for words in the NLQ and words in each of the identified responses, wherein the query specific significance value for a given word in the NLQ is a ratio between the weight value of the given word in the NLQ and a sum of weight values of a set of other words in the NLQ and the query specific significance value for a given word in a given identified response is a ratio between the weight value of the given word in the given identified response and a sum of weight values of a set of other words in the given identified response, such that a query specific significance value of any selected word in any selected NLQ or identified response, is inversely related to a sum of weight values of words, other than the selected word, in the selected NLQ or identified response;
and
score matches between the user NLQ and each of the identified responses by performing, for each match of a given identified response and the user NLQ, a mathematical operation factoring: (i) query significance values of words in the user NLQ also appearing in the individual set of natural language words associated with the given identified response in said data storage; (ii) query significance values of words in the user NLQ synonymous to words appearing in the individual set of natural language words associated with the given identified response in said data storage, and (iii) query significance values of words in the identified responses corresponding to words in the NLQ.
2. The system according to claim 1 , wherein said processing circuitry is further adapted to compare a context in which the user NLQ was submitted to contexts associated with the identified responses, wherein a context in which the user NLQ was submitted is defined as parameters relating to the NLQ other than a text of the NLQ.
3. The system according to claim 1 , wherein said processing circuitry is further adapted to factor degrees of synonimity when assigning weight values to words in the user NLQ, and to words in the individual sets of natural language words, based on rates of occurences of synonyms of each of the words in the at least one knowledgebase.
4. The system according to claim 3 , wherein a degree of synonimity is context dependent.
5. The system according to claim 1 , wherein said processing circuitry is further adapted to factor degrees of synonimity when factoring weight values of words in the user NLQ synonymous to words appearing in an individual set of natural language words associated with a given identified response in said data storage.
6. The system according to claim 5 , wherein a degree of synonimity is context dependent.
7. A system for automatically matching a stored response to a user Natural Language Query (NLQ), said system comprising:
a data storage containing indexed stored responses to natural language queries, each indexed stored response being associated with an individual set of natural language words and indexed based on the individual set of natural language words;
a computing platform including processing circuitry adapted to:
receive digital data representing the user NLQ;
assign weight values to words in the user NLQ and to words in the individual sets of natural language words, based on a rate of occurrence of each of the words in at least one knowledgebase, such that a weight value assigned to a given word is inversely related to a rate of occurrence of the given word in the at least one knowledgebase;
identify responses stored in said data storage having an associated set of natural language words including words appearing in the user NLQ;
identify responses stored in said data storage having an associated set of natural language words including synonyms of words appearing in the user NLQ; and
calculate a query specific significance value for words in the NLQ and words in each of the identified responses, wherein the query specific significance value for a given word in the NLQ is a ratio between the weight value of the given word in the NLQ and a sum of weight values of a set of other words in the NLQ and the query specific significance value for a given word in a given identified response is a ratio between the weight value of the given word in the given identified response and a sum of weight values of a set of other words in the given identified response, such that a query specific significance value of any selected word in any selected NLQ or identified response, is inversely related to a sum of weight values of words, other than the selected word, in the selected NLQ or identified response:
score matches between the user NLQ and each of the identified responses by performing, for each match of a given identified response and the user NLQ, a mathematical operation factoring: (i) query significance values of words in the user NLQ also appearing in the individual set of natural language words associated with the given identified response in said data storage; (ii) query significance values of words in the user NLQ synonymous to words appearing in the individual set of natural language words associated with the given identified response in said data storage, and (iii) query significance values of words in the identified responses corresponding to words in the NLQ.
8. The system according to claim 7 , wherein said processing circuitry is further adapted to compare a context in which the user NLQ was submitted to contexts associated with the identified responses, wherein a context in which the user NLQ was submitted is defined as parameters relating to the NLQ other than a text of the NLQ.
9. The system according to claim 7 , wherein said processing circuitry is further adapted to factor degrees of synonimity when factoring weight values of words in the user NLQ synonymous to words appearing in an individual set of natural language words associated with a given identified response in said data storage.
10. The system according to claim 9 , wherein a degree of synonimity is context dependent.
11. A system for automatically matching a stored response to a user Natural Language Query (NLQ), said system comprising:
a data storage containing indexed stored responses to natural language queries, each indexed stored response being associated with an individual set of natural language words and indexed based on the individual set of natural language words;
a computing platform including processing circuitry adapted to:
receive digital data representing the user NLQ;
assign weight values to words in the user NLQ and to words in the individual sets of natural language words, based on a rate of occurrence of each of the words in at least one knowledgebase and rates of occurences of synonyms of each of the words in the at least one knowledgebase, such that a weight value assigned to a given word is inversely related to a rate of occurrence of the given word in the at least one knowledgebase and is further inversely related to rates of occurrences of synonyms of the given word in the at least one knowledgebase;
identify responses stored in said data storage having an associated set of natural language words including words appearing in the user NLQ;
calculate a query specific significance value for words in the NLQ and words in each of the identified responses, wherein the query specific significance value for a given word in the NLQ is a ratio between the weight value of the given word in the NLQ and a sum of weight values of a set of other words in the NLQ and the query specific significance value for a given word in a given identified response is a ratio between the weight value of the given word in the given identified response and a sum of weight values of a set of other words in the given identified response, such that a query specific significance value of any selected word in any selected NLQ or identified response, is inversely related to a sum of weight values of words, other than the selected word, in the selected NLQ or identified response;
and
score matches between the user NLQ and each of the identified responses by performing, for each match of a given identified response and the user NLQ, a mathematical operation factoring: (i) query significance values of words in the user NLQ also appearing in the individual set of natural language words associated with the given identified response in said data storage, and (ii) query significance values of words in the identified responses corresponding to words in the NLQ.
12. The system according to claim 11 , wherein said processing circuitry is further adapted to compare a context in which the user NLQ was submitted to contexts associated with the identified responses, wherein a context in which the user NLQ was submitted is defined as parameters relating to the NLQ other than a text of the NLQ.
13. The system according to claim 11 , wherein said processing circuitry is further adapted to factor degrees of synonimity when assigning weight values to words in the user NLQ, and to words in the individual sets of natural language words based rates of occurences of synonyms of each of the words in the at least one knowledgebase.
14. The system according to claim 13 , wherein a degree of synonimity is context dependent.