US20050050037A1

US20050050037A1 - Intranet mediator

Info

Publication number: US20050050037A1
Application number: US10/960,804
Authority: US
Inventors: Ophir Frieder; David Grossman
Original assignee: Ophir Frieder; Grossman David Adam
Current assignee: International Business Machines Corp
Priority date: 2001-04-18
Filing date: 2004-10-07
Publication date: 2005-03-03
Also published as: US6904428B2; US20020156771A1

Abstract

An intranet mediator for obtaining direct answers to natural language questions allowing users to search both reconciled structured data repositories and unstructured data sources. The intranet mediator allows the user to obtain an answer to a natural language question without having to surf the data sources in which the answer might be contained, or without being limited to one specific factual item return. The intranet mediator operates on the supposition that most answers to business queries are contained within data sources that have been integrated into the reconciled structured data repository thereby having common schema and known contents. Preselection of the most relevant data source(s) is thus possible before query output. Searches of unstructured data can also be performed for additional context surrounding either the question or the answer. A direct answer is given in answer to the question. If desired, the intranet mediator may also display a list of data sources where additional relevant information may be found.

Description

BACKGROUND OF THE INVENTION

1. Field of the Invention
The present invention relates to a system whereby a direct answer may be given to a specific natural language query through a convenient search of reconciled structured data repositories and unstructured data sources.
2. Discussion of the Related Art
There are two types of digital data gathering commonly in use. One, information retrieval, is concerned with the retrieval of information from unstructured data sources, such as text documents, where each element of the data is not individually defined. The user will enter “search terms” as a data query and the unstructured data will merely be searched for occurrence of these terms. Results of, i.e., the response from, such a search may return the text, i.e., the data, or may, e.g., in a World Wide Web search, only return the location, or site, of the data. The user would then need to read the text or go to each site and locate the occurrence of the search term, which may, or may not, be relevant to an actual question which the user wants answered. This time consuming practice is commonly known as “surfing”.
Information retrieval is thus not geared to efficiently provide a specific answer to a specific question. Attempts to alleviate this problem were the subject of U.S. Pat. No. 6,167,370 to Tsourikov et al., which suggests giving a summary of text findings as a response to a user query. But, for example, when a user wants to know “What are the three best Sushi restaurants in Chicago?” the user does not necessarily care to browse through text summaries, or restaurant guide web sites, which are the likely search results of a known information retrieval search. The surfing in this context may be particularly tedious if the query is submitted to the data sources as an equally weighted string of tokens. For example, where “Chicago” is equally weighted with “Sushi” when figured into the search results, a user may wade through scores of restaurant web sites having nothing to do with Sushi eateries. Avoidance of this problem may require the user to know Boolean logic or other specific search strategy formats, and individually structure each search. The user would most often prefer just a list of three Sushi restaurants in Chicago in answer to this natural language question.
The second type of digital data gathering commonly in use is the structured data source search, where structured data within one or more specific structured data sources or reconciled structured data repositories, usually privately owned and accessed, are searched to return a specific answer. “Reconciled” as used herein refers to reconciling data to a common schema of a structured data storage entity. Techniques such as extracting, cleaning, transforming, and loading the data are used to place the data into a data repository with a common schema, here the “reconciled structured data repository.” As is known, a reconciled structured data repository may contain unstructured data, e.g., pages of text, video clips, etc., and such unstructured data are reconciled by known techniques, including those listed above, to provide a structured means by which to search, access, and potentially retrieve them within the reconciled structured data repositories.
One known structured data storage technique for use with reconciled structured data repositories is a “data warehouse,” an example of which is described in Hellerstein et al., Independent, Open Enterprise Data Integration, Bulletin of the IEEE Computer Society Technical Committee on Data Engineering, 1999. As will be known to those in the art, a “physical data warehouse” can be at least one, or a collection of, reconciled structured data repositories of data. Thus a “physical data warehouse” is also a repository of integrated, proprietarily loaded and controlled data sources assumed to be under the control of a single entity. As per the Hellerstein article, the “physical data warehouse” may be constructed according to extraction/transform/load (ETL) techniques where the selected data are first extracted, i.e., selected for retention and cleaned up to retain those portions necessary for the common schema, then transformed, i.e., integrated within the structure of the reconciled structured data repository or physical data warehouse format by reconciling the data to the common schema of the structured data repository, and then loaded under the common schema into the reconciled structured data repository. As will be known to those of ordinary skill in the art, a reconciled structured data repository may also store unstructured data that have been reconciled to the common schema of the reconciled structured data repository by having or pointer to identifying metadata for the unstructured data. Thus, the structured and unstructured data within the reconciled structured data repository can be retrieved by structured search schemes. While some art references predating the current nomenclature and usage have labeled a mere aggregation or collection of unstructured databases searched by information retrieval means and terms as “data warehouses,” such unstructured databases do not contain reconciled structured data repositories, do not access their data in a structured manner, and are not “data warehouses” as the term is now known to those in the art because the data therein is not transformed to a common schema.
In the past, the nonintegrated data sources were required to be searched one data source at a time. Integration of their individual data sources is generally performed by private business to enable answers to queries whose answers require more than one factual component. This integration is expensive and can remain underutilized for reasons such as an arcane nature of query formulation or because extensive data source knowledge may be required of the user to make a rational search selection. That is, the user may need to know where to look and how to look to expect a relevant answer. Concurrent searching of nonintegrated structured data sources, and merging of their results, to solve some of these problems, was the subject of U.S. Pat. No. 5,995,961 to Levy, et al.
Further, additional information, beyond the specific factual components of a query, cannot be provided from the results of a data source search. For example, assume the data source user, or searcher, wishes to know the building on the I.I.T. campus with the largest number of rooms. The user cannot expect a picture of the building, or a link to a picture of the building, returned with the search results, even though the user might wish to see such a picture.
U.S. Pat. No. 6,078,924 to Ainsbury et al. illustrates a technique of digital data gathering. According to this patent, the user is allowed to aggregate data found in the user's previous searches on a specific topic into a central file. This central file can then be controlled from a commercial desktop computer application to facilitate searching of the data.
What is needed in the art is a system whereby the user can take advantage of both information retrieval and reconciled structured data repository data types of digital data gathering concurrently to provide a direct answer to a specific question, and preferably provide further context for that answer. It is also desirable that the query be accepted in a natural language format whereby the user needs no special skills in query formulation. It is further desirable that the query be intelligently parsed so as to weight the relevant parts of the query and that synonyms of the natural language query be provided to give a more thorough search and accurate answer. It is further desirable that the answers, and any related information, be limited in number to only that required or most relevant to the query.

Definitions

“Query” refers herein to any form of searchable subject matter, and may include query tokens, or elements of a total query, whether aggregate or separate, unless otherwise limited or defined by the context of the disclosure.
“Data” refers herein to any form of digitally stored information, unless otherwise limited or defined by the context of the disclosure.
“Concurrent” means within the time frame between question output and answer display and does not necessarily imply that searches happen simultaneously.
“Direct answer” and “most likely answer” are used interchangeably herein and refer to the best available answer, whether factually based, referencing additional data, or refusal to answer, based upon the results of the data retrieved by the searches of the intranet mediator. A direct answer is to be distinguished from an information retrieval type “response” which merely indicates the presence of a document containing the queried term.
“Common schema” means an organizational definition of structured data shared by multiple data sources.
“Data source” means a logically, independently operating data storage, search, retrieval, and manipulation system. The system may store data of any digital form.

SUMMARY OF THE INVENTION

The intranet mediator of the present invention obtains for the user a direct, or most likely, answer to a natural language question. The intranet mediator provides the user with a search of both structured data sources and a repository of unstructured data sources. The reconciled structured data repository data sources will preferably be integrated to provide for ease of searching a rapid attainment of search results. The reconciled structured data repository is previously constructed, e.g., by the corporate owner thereof, to integrate and abstract its collections of data sources with common schema, and providing meta-data within the reconciled structured data repository. The meta-data will give a global overview of the reconciled structured data repositories thus making the repositories easily searchable. The unstructured data sources may be data such as internal video, audio, or document files stored within the private databases of the owner, or they may be public sources, e.g., collections of documents available via the worldwide web or the like, or both. The unstructured data sources may be provided with a meta-data repository also.
The intranet mediator allows the user to input, and obtain an answer to, a natural language question without having to read text or surf the data sources in which the answer might be contained, or without being limited to one specific factual item return. The intranet mediator operates in part on the supposition that most answers to a businessperson's questions are contained within privately owned data repositories of the business that were likely to be already integrated into a reconciled structured data repository. Selection of the most relevant data sources prior to searching is therefore possible. The direct answer selection to the user's question is accordingly weighted to the search answer from these reconciled structured data repositories. Searching of the unstructured data sources may be performed automatically, or upon determination that answers are not likely from the structured data sources, or if additional context surrounding either the query or the answer is justified or desired. A direct, or most likely, answer is then given to the user in answer to the input question. If desired, the intranet mediator may also display a list of related data or data sources where additional information relevant to the user's question may be found.
The intranet mediator desirably includes the logical functionality of a natural language question input module; a parser module for assembling a search query from the natural language question; a query expander module; an unstructured data source manager operably connected to at least one unstructured data source; a data source selection module operably connected to a meta-data repository; a dispatcher module interfacing between the data source selection module and both a structured data source manager and the unstructured data source manager, the structured data source manager being operably connected to a reconciled structured data repository; a results manager module operably connected to both of the data source managers; and an answer output module.
Discussion of the modules will be given herein with respect to specific functional tasks or task groupings that are in some cases arbitrarily assigned to the specific modules for explanatory purposes. It will be appreciated by the person having ordinary skill in the art that an intranet mediator according to the present invention may be arranged in a variety of ways, or that functional tasks may be grouped according to other nomenclature or architecture than is used herein without doing violence to the spirit of the present invention.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG. 1 is a block diagram representing the construction of a reconciled structured data repository such as might be used according to one embodiment of the current invention.
FIG. 2 is a block diagram representing the architecture of an intranet mediator according to one embodiment of the current invention.

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENT

Referencing the block diagram of FIG. 1, as is known, raw data intended for placement into a reconciled structured data repository 13 may placed in structured form as at box 3, or may be left unstructured as at line 5. Both structured and unstructured data are then reconciled, as at box 7, to the common schema of the reconciled structured data repository 13 and then loaded, i.e., stored, as at line 9 into a reconciled structured data repository 13 for future search, accessing, and retrieval, such as discussed in conjunction with the present invention below.
Referencing the block diagram of FIG. 2, the preferred embodiment of an intranet mediator 11 according to the present invention comprises means for the searching of both structured data sources, such as a multiplicity of databases integrated into a reconciled structured data repository 13, e.g., a data warehouse, and unstructured data sources, collectively 15, whether in the owner's proprietary databases or within a public collection such as the Internet, to arrive at a direct answer to a natural language question input by the user.
A user interface 17, such as a known graphical user interface, accepts input of a user question in natural language format and may allow the user to manually select data sources if desired. As indicated by the two boxes in FIG. 2 labeled 17, the user interface 17 is preferably a part of an input/output user interface that is also tasked with displaying an answer to the user question.
A parser module 19 then accepts the natural language question input by the user, as at line 20, parsing and assembling the relevant concepts of the natural language question into a query, or queries, of weighted search tokens, hereinafter referred to simply as a query, and also eliminating the irrelevant, or non-indicative, words of the natural language question from use as search tokens. For example, in the query: “What are the three best Sushi restaurants in Chicago?,” words irrelevant to data retrieval such as “are” and “the” may be eliminated from use as tokens by the parser module. Various techniques such as grammar matching, lattices, partial lattices, etc. for accomplishing the tasks of the parser module 19 are available and considered within the skill of the person having ordinary skill in the art.
A query expander module 21 is operably connected to the parser module 19, as at line 22, for receiving the parsed query tokens and obtaining additional terms synonymous or analogous to the tokens and expanding the query with additional tokens if desirable to expand the chance of a return of information relevant to the query. Various techniques such as thesauri, dictionaries, term expansion, stemming, phrase generation, and the like are available to accomplish this query, or token, expansion and are considered within the skill of the person having ordinary skill in the art.
The query expander module 21 then passes the tokens, preferably including the expanded tokens, to an unstructured data source manager 23, as at line 24, for a limited use specialty search to acquire additional synonyms, or terms analogous to, the query tokens and expanded query tokens, if any. The unstructured data source manager 23 initiates a search, as at lines 26, of one or more repositories of unstructured data sources 15, hereinafter referred to in the singular for ease of explanation, and obtains the search results therefrom, as at lines 28, with any acceptable method of information retrieval. Additional tokens, if any, may be filtered from the results of the unstructured data source search returned to the query expander module 21 from unstructured data source manager 23 via line 46 and added to the expanded query for return to the parser module 19 through the query expander module 21, as at line 30. The unstructured data repository 15, or an operable link to such a repository, may also be considered as a part of the intranet mediator 11 according to certain embodiments of the present invention. Unstructured data sources may include private or public data repositories, e.g., a private video archive and the World Wide Web, respectively.
The parser module 19 will then pass the expanded query to the data source selection module 25, as at line 32. The data source selection module 25 determines the most likely data source or sources to contain an answer for each token of the expanded query; e.g., adaptive, heuristic, hard-coded, user-directed, standard, handwritten, or data-mined inquiry techniques through its connection, as at line 34, to a meta-data repository 27 abstracting the contents of a data warehouse 13 of structured data sources. Meta-data for the unstructured data repository 15 may further be contained in the meta-data repository 27 in certain embodiments of the present invention. The meta-data repository, or an operable link to such a repository, may also be considered as a part of the intranet mediator according to certain embodiments of the present invention.
Upon selection of the appropriately likely data sources to return an answer for each token of the expanded query, the data source selection module 25 will pass the data source selections and accompanying search tokens onto the dispatcher module 31, as at line 36. The dispatcher module 31 will then route each token to be searched to the appropriate data source manager 33, 23. Routing will commonly occur via an intranet although other networks including the Internet or simply direct routing are also within the scope of this invention, as at line 38 to the structured data source manager 33 or to the unstructured data source manager 23 via line 40. The structured data source manager 33 and the unstructured data source manager 23 will then interface most efficiently with the reconciled structured data repository 13 and unstructured data sources 15, via lines 42 and 26, respectively, according to the individual requirements of the data sources. The reconciled structured data repository 13, e.g., may be a physical data warehouse of integrated and structured private data sources, and may be constructed according to known techniques such as extraction/transform/load (ETL) techniques. One of ordinary skill in the art will recognize that other reconciled structured data repository development techniques are likewise within the scope of this invention. The reconciled structured data repository 13, or an operable link to such a reconciled structured data repository, may also be considered a part of the intranet mediator 11 according to some embodiments of the present invention.
Search results are returned to the structured and unstructured data source managers, 33, 23 respectively, via lines 44, 28, respectively, for additional processing if any, and then forwarded via lines 46, 48 respectively to the results manager module 51.
The results manager module 51 desirably accepts and consolidates the results of the structured and unstructured data source searches for each search token and integrates the results of the searches. Duplicate results are also eliminated as appropriate. The results manager module 51 then weights or ranks the results and at least one most likely answer is selected. In the presently preferred embodiment, if an answer is returned from the structured data source, for example, the data warehouse, it is published as the most likely answer. It will be understood that selection of the direct answer from the unstructured data source is within the scope of this invention. In the described embodiment, the associated data links, extracted text, or the like, from the unstructured data source search may be further selected and ranked. An answer module portion 53 of the user interface 17 may then receive answer data via line 50 and may then perform any formatting necessary for the display via the user interface 17, of the most likely answer and, if desired, the associated data or data links available as additional context for the direct answer.
The user interface 17 of the described embodiment will wait for some specified time interval or specified number of results to accumulate and then display the currently ranked results. The user interface 17 will further continue receiving results and ranking them in conjunction with the total accumulated results to capture answers which may have been delayed for various reasons. The user is then preferably presented with an option to activate a “get next” display command, and the now possibly revised rankings or additionally aggregated results are displayed.
Having thus described an intranet mediator for searching both structured and unstructured data sources for arriving at a most likely answer to a natural language question input by the user; it will be appreciated that many variations thereon will occur to the artisan upon an understanding of the present invention, which is therefore to be limited only by the appended claims.

Claims

1. A method for digital data gathering in answer to a query, comprising: conducting concurrent searching of a reconciled structured data repository and unstructured data sources, and preselecting at least one data source from the reconciled structured data repository and the unstructured data sources most likely to contain a direct answer to the query before submitting the query to the data sources.

2. The method for digital data gathering in answer to a query, according to claim 1, further comprising: combining results from said reconciled structured data repository and unstructured data source searches.

3. The method for digital data gathering in answer to a query, according to claim 1, further comprising: performing a search of the query in the preselected data sources, combining results from the preselected data source searches, sorting the results and providing a direct answer to the query.

4. The method for digital data gathering in answer to a query, according to claim 1, further comprising: combining a plurality of structured data sources into a reconciled structured data repository with a meta-data repository.

5. The method for digital data gathering in answer to a query, according to claim 4, further comprising: performing a search of the query in the preselected data sources, combining results from the preselected data source searches, sorting the results and providing a direct answer to the query.

6. A method of digital data gathering for providing a direct answer to a natural language question, comprising:

a) accepting input of a natural language question;

b) identifying the relevant concepts of the natural language question;

c) assembling the relevant concepts of the natural language question into a query;

d) identifying a reconciled structured data repository likely to contain an answer to the query;

e) performing a first search of the query in the reconciled structured data repository;

f) performing a second search of the query in an unstructured data source;

g) integrating the results of the first and second searches and selecting a direct answer to the natural language question; and

h) displaying the direct answer to the natural language question.

7. The method of digital data gathering for providing a direct answer to a natural language question, according to claim 6, further comprising: combining a plurality of structured data sources into the reconciled structured data repository.

8. The method of digital data gathering for providing a direct answer to a natural language question, according to claim 6, further comprising: combining a plurality of unstructured data sources into the reconciled structured data repository.

9. The method of digital data gathering for providing a direct answer to a natural language question, according to claim 6, further comprising: combining a plurality of structured and unstructured data sources into the reconciled structured data repository.

10. A method for digital data gathering in answer to a query, comprising:

conducting concurrent searching of a reconciled structured data repository and unstructured data sources,

preselecting individual structured data sources and unstructured data sources most likely to contain a direct answer to the query before submitting the query to the data sources; and

sorting results of the data source searches and providing a direct answer to the query.

11. A method for digital data gathering in answer to a query, comprising:

combining data sources into a reconciled structured data repository with a meta-data repository,

conducting a search of the data within the reconciled structured data repository,

conducting a search of unstructured data sources outside of the reconciled structured data repository if a direct answer is not selected from the reconciled structured data repository search, and

sorting results of the search and providing a direct answer to the query.

12. The method according to claim 1 further including conducting a search of unstructured data sources outside of the reconciled structured data repository if a direct answer is not selected from the reconciled structured data repositories search.

13. The method according to claim 1 further including conducting a search of an unstructured data source outside of the reconciled structured data repository and combining and sorting the results of the unstructured data source search outside of the reconciled structured data repository with the results of the reconciled structured data repository search.

14. The method according to claim 12 wherein the selection of the direct answer is weighted to the search results from the reconciled structured data repository.

15. The method according to claim 13 wherein the selection of the direct answer is weighted to the search results from the reconciled structured data repository.