Visualization of information using graphical representations of context vector based relationships and attributes

• US 5,794,178 A
• Filed: 04/12/1996
• Issued: 08/11/1998
• Est. Priority Date: 09/20/1993
• Status: Expired due to Term

• Current Assignee: Fair Isaac Corporation
• Original Assignee: HNC Software Inc. (Fair Isaac Corporation)
• Inventors: Carleton, Joel Lawrence, Caid, William Robert
• Primary Examiner(s): Hudspeth, David R.
• Assistant Examiner(s): Thomas, Joseph
• Application Number:
  US08/632,519
• Time in Patent Office: 851 Days
• Field of Search: 704/9, 704/10, 395/22, 395/759, 395/760, 707/5, 707/531, 707/532, 345/440, 345/334, 345/339, 345/349
• US Class Current: 704/9
• CPC Class Codes: G06F 16/3347 using vector based model
• CPC Class Codes: G06F 16/338 Presentation of query results
• CPC Class Codes: G06F 16/58 Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually
• CPC Class Codes: Y10S 707/99932 Access augmentation or optimizing
• CPC Class Codes: Y10S 707/99935 Query augmenting and refining, e.g. inexact access

Abstract

A system and method for generating context vectors for use in storage and retrieval of documents and other information items. Context vectors represent conceptual relationships among information items by quantitative means. A neural network operates on a training corpus of records to develop relationship-based context vectors based on word proximity and co-importance using a technique of "windowed co-occurrence". Relationships among context vectors are deterministic, so that a context vector set has one logical solution, although it may have a plurality of physical solutions. No human knowledge, thesaurus, synonym list, knowledge base, or conceptual hierarchy, is required. Summary vectors of records may be clustered to reduce searching time, by forming a tree of clustered nodes. Once the context vectors are determined, records may be retrieved using a query interface that allows a user to specify content terms, Boolean terms, and/or document feedback. The present invention further facilitates visualization of textual information by translating context vectors into visual and graphical representations. Thus, a user can explore visual representations of meaning, and can apply human visual pattern recognition skills to document searches.

Claims

1. A computer-implemented process of providing a visual representation of information content for a record in a computer system including a display device for displaying visual information at positions on a display screen, the record having a stored summary vector, the process comprising:

(a) defining at least two axes with respect to the display screen of the display device, each axis being associated with a coordinate term having a coordinate context vector;

(b) determining a position on the display screen of the display device for a visual representation of the record relative to each axis by dot product calculation of the summary vector of the record with the coordinate context vector of the axis; and

(c) displaying the visual representation of the record at the determined position on the display screen relative to each axis.

2. The computer-implemented process of claim 1, further comprising:

(b.1) defining at least one attribute term representing a visual attribute of the visual representation and having a context vector; and

(b.2) determining a value for the visual attribute by dot product calculation of the summary vector of the record with the context vector of the attribute term;

and wherein step (c) comprises displaying the visual representation according to the determined value for the visual attribute at the determined position.

3. The computer implemented process of claim 1, wherein step (a) comprises:

(a.1) receiving a query including a plurality of terms;

(a.2) determining a context vector for each term;

(a.3) selecting at least two of the terms as coordinate terms;

(a.4) defining an axis for each selected coordinate term relative to the display screen.

4. The computer implemented process of claim 3 further comprising:

(d) selecting a plurality of records from a database according to the dot product of the summary vector of the record and the context vectors of the terms; and

(e) repeating steps (b) and (c) for each of the selected plurality of records, to create a visual representation of each of the selected plurality of records at positions on the display screen.

5. The computer implemented process of claim 1, wherein there are three pseudo-orthogonal display axes each associated with a context vector, and each visual representation is a pseudo-three dimensional representation positioned with respect to each axis according to a dot product of the summary vector of the record and the context vector associated with the axes.

6. A computer-implemented process of providing a visual representation of information content for a record in a computer system including a display device for displaying the visual representation, the display device and visual representation having modifiable visual attributes that modify the display of the visual representation, the record having a stored summary vector, the process comprising:

(a) defining at least two visual attributes, each visual attribute being associated with a context vector and having a plurality of potential values applicable to the visual representation of the record as displayed on the display device;

(b) for each visual attribute, determining a value of the visual attribute to be applied to the visual representation of the record on the display device by a dot product calculation of the summary vector of the record with the context vector of the visual attribute; and

(c) displaying the visual representation of the record on the display device with the determined values of each of the visual attributes.

7. The computer implemented process of claim 6, further comprising:

(d) repeating steps (b) and (c) for each of a plurality of records, each having a summary vector, to create a visual representation for each of the plurality of records.

8. The computer implemented process of claim 7, further comprising:

(e) receiving a query including a plurality of terms;

(f) determining a context vector for selected terms of the query;

(g) for each selected term;

(g.1) determining for each visual attribute, a value of the visual attribute to be applied to a visual representation of the term on the display device by a dot product calculation of the context vector of the term with the context vector of the visual attribute; and

(g.2) displaying the visual representation of the term on the display device with the determined value of the visual attribute.

9. A computer-implemented process of providing a visual representation of information content for a record in a computer system including a display device for displaying the visual representation, the display device and visual representation having modifiable visual attributes that modify the display of the visual representation, the record having a stored summary vector having a plurality of components, the process comprising:

(a) defining at least one attribute term representing a visual attribute applicable to the visual representation of the record as displayed on the display device, the visual attribute having a context vector;

(b) determining a value for the visual attribute applicable to the visual representation by dot product calculation of the summary vector with the context vector of the attribute term; and

(c) displaying a visual representation of the record on the display device according to the determined value of each visual attribute.

10. A computer readable media storing thereon a program product executable by a processor for providing a visual representation of information content for a record in a computer system including a display device for displaying visual representations at positions on a display screen, the display device and visual representations having modifiable visual attributes that modify the display of the visual representations, the record stored in a memory and having a summary vector having a plurality of components, the program product comprising:

an axis definer, executable by the processor for defining at least two axes with respect to the display screen, each axis being associated with a coordinate term having a coordinate context vector, and for storing data defining the axes in the memory;

a position determiner, executable by the processor, for reading the axis data and determining a position on the display screen for a visual representation of the record relative to the axes of the display screen by comparing the summary vector of the record with the coordinate context vector with each of the axis vectors, and for storing the position in the memory; and

a display engine, executable by the processor and coupled to the display device, for reading the position and displaying the visual representation of the record at the determined position on the display screen.

11. The computer readable media of claim 10, further comprising:

attribute definer, executable by the processor, for defining at least one attribute term representing a visual attribute applicable to a visual representation of a record on the display screen, the attribute term having a context vector; and

a value determiner, executable by the processor, for determining a value for the visual attribute by comparing the summary vector with the context vector of the attribute term;

and wherein the display engine displays the visual representation of the record on the display screen according to the determined value for the visual attribute.

12. A computer readable media storing thereon a program product executable by a processor for providing a visual representation of information content for a record, the record stored in memory and having a summary vector having a plurality of components, the program product comprising:

an attribute definer, executable by the processor, for defining at least one attribute term representing a visual attribute and having a context vector;

a value determiner, executable by the processor, for determining a value for the visual attribute by comparing the summary vector with the context vector of the attribute term; and

a display engine, executable by the processor, for displaying the visual attribute according to its determined value.

13. In a computer with a processor and memory, a computer-implemented method of determining a similarity between a first data item and a second data item in a corpus of data items stored in the memory, the method comprising:

defining a set of information elements that represent the data items in the corpus, the information elements having a measurable proximity between each pair of information items, each information item having a frequency of occurrence in the corpus;

generating, via the processor, a context vector associated with each information element as a function of the proximity of each information element with respect to other information elements in the corpus, and the frequency of occurrence of each information element in the corpus;

generating, via the processor, a first summary vector associated with the first data item as a function of the context vectors associated with the information elements that represent the first data item;

generating, via the processor, a second summary vector associated with the second data item as a function of the context vectors associated with the information elements that represent the second data item; and

determining a similarity between the first and the second data items by comparing the first and the second summary vectors.

14. The computer-implemented method of claim 13 wherein:

there are N information elements in the set of information elements used to represent the data items in the corpus;

R is an N×

N positive definite mutual co-occurrence matrix of the N information elements;

W defines a set of n N×

1 context vectors associated with each information element, wherein W is defined by;

space="preserve" listing-type="equation">W=D.sup.1/2 Q.sup.T

where Q^T is the transpose of an orthogonal matrix with unit-length eigenvectors of R for columns; and

,

D is a diagonal matrix with eigenvalues of R as its diagonal entries.

15. The computer-implemented method of claim 14 further comprising:

determining the eigenvalues and eigenvectors of R with an adaptive neural network.

16. The computer-implemented method of claim 14 further comprising:

reducing the dimension of W to a set of N N×

1 context vectors by singular value decomposition, wherein n is less than N.

17. In a computer comprising a processor, a computer-implemented method for associating a first data domain comprising a plurality of first information elements with a second data domain comprising a plurality of second information elements to form a universal meaning space to provide a meaning for second information elements from a known meaning of first information elements, the first information elements having measurable proximity between any pair of first information elements, each first information element having a frequency of occurrence in the first data domain, the second information elements having a measurable proximity between any pair of second information elements, each second information element having a frequency of occurrence in the second data domain, the method comprising the steps of:

generating, via the processor, a first context vector associated with each first information element in the first data domain as a function of the proximity of each first information element with respect to other first information elements in the first data domain and the frequency of occurrence of each first information element;

receiving a user put identifying first information elements in the first data domain that are substantially equivalent in meaning to second information elements in the second data domain;

associating the first context vectors of each of the identified first information elements with the corresponding substantially equivalent second information elements;

generating, via the processor, a second context vector associated with each second information element in the second data domain that does not have an associated context vector, wherein the second context vector is generated as a function of the proximity of each second information element with respect to other second information elements in the second data domain and the frequency of occurrence of each second data element; and

storing the first and second context vectors associated with the first and second information elements in the computer storage.

18. The computer-implemented method of claim 17, wherein the first data domain is a first human language comprising word information elements of the first human language, and the second data domain is a second human language comprising word information elements of the second human language, wherein the user input identifies word information elements in the first human language that have substantially equivalent meaning to word information elements in the second human language.

19. The computer-implemented method of claim 17, wherein the first data domain is English comprising English word information elements, and the second data domain is a non-English language comprising non-English word information elements, wherein the user input identifies English word information elements that have substantially equivalent meaning to non-English word information elements in the non-English language.

20. The computer-implemented method of claim 17, wherein the first data domain is text including word information elements, and the second data domain is an image database including image information elements, wherein the user input identifies word information elements that have substantially equivalent meaning to image information elements.

21. The computer-implemented method of claim 17, wherein the first data domain is a video database including spatio-temporal information elements, and the second data domain is an audio database including temporal audio information elements.

22. The computer-implemented method of claim 17, wherein the method is performed in a computer-based gisting system that determines a meaning of a set of second information elements in the second data domain by selectively displaying first information elements having first context vectors similar to the second context vectors of the set of second information elements.

23. The computer-implemented method of claim 17, further comprising the steps of:

receiving an input including a set of first information elements in the first data domain;

comparing the first context vectors associated with the set of first information elements with second context vectors associated with second information elements in the second data domain to determine a similarity measure between the first and second context vectors;

selecting a set of second context vectors and associated second information elements from the second data domain as a function of the similarity measure; and

displaying the selected set of second information elements.

24. The computer implemented method of claim 23, wherein the similarity measure is a dot product between context vectors.