System for retrieving multimedia information from the internet using multiple evolving intelligent agents

• US 6,304,864 B1
• Filed: 04/20/1999
• Issued: 10/16/2001
• Est. Priority Date: 04/20/1999
• Status: Active Grant

• Current Assignee: TextWise Company, LLC
• Original Assignee: TextWise Company, LLC
• Inventors: Liddy, Elizabeth D., Yu, Edmund Szu-li
• Primary Examiner(s): Powell, Mark R.
• Assistant Examiner(s): Holmes, Michael B.
• Application Number:
  US09/295,190
• Time in Patent Office: 910 Days
• Field of Search: 707/4, 706/20, 706/25
• US Class Current: 706/15
• CPC Class Codes: G06F 16/9535 Search customisation based on user profiles and personalisation
• CPC Class Codes: G06N 3/086 using evolutionary algorithms, e.g. genetic algorithms or genetic programming

Abstract

A system for retrieving multimedia information is provided using a computer coupled to a computer-based network, such as the Internet, and particularly the World Wide Web (WWW). The system includes a web browser, a graphic user interface enabled through the web browser to allow a user to input a query representing the information the user wishes to retrieve, and an agent server for producing, training, and evolving first agents and second agents. Each of the first agents retrieves documents (Web page) from the network at a different first network address and at other addresses linked from the document at the first network address. Each of the second agents executes a search on different search engines on the network in accordance with the query to retrieve documents at network addresses provided by the search engine. The system includes a natural language processor which determines the subject categories and important terms of the query, and of the text of each agent retrieved document. The agent server generates and trains an artificial neural network in accordance with the natural language processed query, and embeds the trained artificial neural network in each of the first and second agents. During the search, the first and second agents process through their artificial neural network the subject categories and important terms of each document they retrieve to determine a retrieval value for the document. The graphic user interface displays to the user the addresses of the retrieved documents which are above a threshold retrieval value. The user manually, or the agent server automatically, selects which of the retrieved documents are relevant. Periodically, the artificial neural network of the first and second agents is expanded and retrained by the agent server in accordance with the selected relevant documents to improve their ability to retrieve documents which may be relevant to the query. Further, the agent server can evolve an artificial neural network based on the current artificial neural network, the retrieved documents, and their selected relevancy, by iteratively producing, training, and testing several generations of neural networks to produce an evolved agent. The artificial neural network of the evolved agent then replaces the current artificial neural network used by the agents to search the Internet. One or more concurrent search of the Internet may be provided.

Claims

1. A system for retrieving information on a computer coupled to a computer-based network, such as the Internet, in accordance with a query representing the information a user wishes to retrieve, said system comprising:

means for producing a plurality of first agents and second agents in which said first agents each retrieve documents at a different first network address and at other addresses linked from the document at the first network address, and said second agents each execute a search on a different search engines via the network in accordance with said query and retrieves documents at network addresses provided by the executed search;

said first and second agents each comprising an artificial neural network trained in accordance with said query for determining for each of the retrieved documents by said agents a retrieval value representing the proximity of the content of the retrieved documents to said query; and

means for displaying to the user the addresses of the retrieved documents above a threshold retrieval value.

2. The system according to claim 1 further comprising means for enabling said user to input said query.

3. The system according to claim 1 further comprising a natural language processor for determining subject categories and terms representative of said query, and means for generating and training said artificial neural network having inputs in accordance with said subject categories and terms representative of said query.

4. The system according to claim 3 wherein said natural language processor operates on each of the retrieved documents from said first and second agents to determine the subject categories and terms representative of the retrieved document, and each of said first and second agents set the inputs of the artificial neural network of the agent in accordance with the subject categories and terms representative of each of the retrieved documents by the agent to determine the retrieval value of the retrieved document.

5. The system according to claim 1 further comprising means for selecting which ones of said retrieved documents are the relevant to said query.

6. The system according to claim 1 further comprising means for expanding the artificial neural network of said first and second agents in accordance with the frequency of subject categories and terms present in said selected relevant documents, and training said artificial neural network of said first and second agents in accordance with training patterns based upon said selected relevant documents.

7. The system according to claim 6 wherein said expanding and training means is enabled periodically.

8. The system according to claim 1 further comprising means for evolving the artificial neural network of said first and second agents in accordance with said retrieved documents by said agents and said selected relevant documents, in which multiple generations of third agents are generated having artificial neural networks with subcombination of the input of the artificial neural network of said first and second agents and each successive generation of third agents have a higher chance of obtaining inputs of artificial neural networks of third agents of the previous generation which performed best at classifying a group of said retrieved documents as relevant and non-relevant as provided by said selecting means.

9. The system according to claim 1 further comprising means for enabling said user to select one or more retrieved documents on said displaying means as relevant.

10. The system according to claim 1 further comprising means for automatically selecting the relevant retrieved documents.

11. The system according to claim 1 further comprising:

a Web browser; and

a graphical user interface enabled through the Web browser for said user to input said query and information characterizing the type of documents to be retrieved, wherein said query and said information represent a user search profile.

12. The system according to claim 11 further comprising an agent server for receiving said user search profile and generates an agent leader in accordance with said user search profile responsible for enabling said first and second agent producing means.

13. The system according to claim 12 wherein said agent server responsive to receiving multiple different ones of user search profiles generates multiple different agent leaders in accordance with each of said user search profiles, wherein each of the agent leader are responsible for enabling said producing means to provide a different group of said first and second agents under each of the agent leaders.

14. The system according to claim 11 further comprising a database and a database server for storing at least said user profile, information representing said artificial neural network of said first and second agents, and results of the retrieved documents in terms of at least their network addresses.

15. The system according to claim 1 further comprising means for generating and training an artificial neural network common to each of said first and second agents in accordance with said query.

16. The system according to claim 1 further comprising means for enabling said user to select at least one of said first network addresses.

17. The system according to claim 1 wherein at least one of said addresses of documents retrieved by said second type of agents provides one of said first network address.

18. A method for retrieving information on a computer coupled to a computer-based network, such as the Internet, in accordance with a query representing the information a user wishes to retrieve, said method comprising the steps of:

producing a plurality of first agents and second agents in which said first agents each retrieve documents at a different first network address and at other addresses linked from the document at the first network address, and said second agents each execute a search on a different search engines via the network in accordance with said query and retrieves documents at network addresses provided by the executed search;

generating a trained artificial neural network common to each of said first and second agents in accordance with said query for determining for each of the retrieved documents by said agents a retrieval value representing the proximity of the content of the retrieved documents to said query; and

displaying to the user the addresses of the retrieved documents above a threshold retrieval value.

19. The method according to claim 18 further comprising the steps of:

selecting which of said displayed addresses of the retrieved documents are relevant; and

periodically revising and training said artificial neural network of said first and second agents in accordance with said selected retrieved documents.

20. The method according to claim 18 further comprising the step of:

evolving said artificial neural network of said first and second agents in accordance with said retrieved documents and said selected retrieved documents.

21. The method according to claim 18 further comprising means for enabling said user to select at least one of said first network addresses.

22. The method according to claim 18 wherein at least one of said addresses of documents retrieved by said second type of agents provides one of said first network addresses.

23. A system for retrieving information from the Internet utilizing multiple intelligent agents comprising:

a computer system having a graphical user interface to input a query, means for accessing the Internet, means for producing a plurality of agents in which each of said agents retrieves documents at a first address on the Internet and at other addresses linked to the document at the first address, and means for determining the subject and important terms of the text of the query and of each the documents retrieved;

each of said agents having a common neural network for determining the relevancy of each of the document retrieved by the agent, said neural network having a plurality of inputs and an output in which said inputs are based upon the subject and important terms of the query and said output representing a relevance value of each of the documents applied to the neural network;

said computer system having means for training the neural network in accordance with the query, means for selecting which of said retrieved documents are relevant, means for periodically updating the neural network with additional inputs based on the subject and important terms of the selected retrieved documents and training the updated neural network using said selected retrieved documents to provide a retrained neural network for each agent, and means for evolving and training a plurality of different neural networks in which each is based on subset of the inputs of the retrained neural network, and iteritively evolving and training a new set of different neural networks having a subset of the inputs of such evolved trained different neural networks which best classifies documents as relevant until one of the evolved neural networks is the best classifier of documents as relevant to provide said one neural network for each agent; and

said graphic user interface displaying to a user the results of the documents retrieved by said first and second agents.

24. A method for neurogenically evolving a parent artificial neural network having a plurality of inputs each characteristic of a different feature using multiple sets of one or more of said training features in which each set has a classification of a plurality of known classes, said method comprising the steps of:

generating a plurality of agents each having an artificial neural network with a different subset of said features of said parent artificial neural network;

dividing said multiple sets into a training group and a test group;

training the artificial neural network of each of said agents with said multiple sets of said training group;

testing the artificial neural network of each of said agents using said multiple sets of said test group to determine the number of sets correctly classified;

determining a fitness function for each of said agents the number of sets correctly classified by the total number of sets in said test group;

rank the agents by their fitness function;

generating a plurality of next generation agents each having an artificial neural network with a different subset of said features of said parent artificial neural network in which said artificial neural network of the next generation agents has a greater chance of including said features of the artificial neural network of said higher ranked agents of the prior generation;

repeating said training step, testing step, determining step, identifying step and said step of generating a plurality of next generation agents in accordance with said next generation of agents until one of a maximum number of generation of agents have been produced, and two successive generations of agents each having identical features are produced, in which the agent having the highest fitness function represents an evolved artificial neural network.