Data processing: database and file management or data structures – Database design – Data structure types
Reexamination Certificate
2000-05-02
2004-06-15
Robinson, Greta (Department: 2177)
Data processing: database and file management or data structures
Database design
Data structure types
C706S055000, C707S793000
Reexamination Certificate
active
06751621
ABSTRACT:
BACKGROUND OF THE INVENTION
1. Technical Field
The present invention relates to information analysis and, more particularly, to a semantic representation of information and analysis of the information based on its semantic representation.
2. Description of the Related Art
The ever-increasing demands for accurate and predictive analysis of data has resulted in complicated processes that requires massive storage capacity and computational power. The amount and type of information required for different types of analysis can further vary based on the required results. Oftentimes, it is necessary to filter the required information from a storage system in order to perform the desired analysis. One method of storing information is through the use of relational database tables. A specific location is designed for high capacity storage and used to maintain the information. Currently, the location can be local or off-site. Regardless of the location, various types of network and intemetworking connections (i.e., LAN, WAN, Internet) can be used to access the information.
The most common method of accessing and filtering information is through the use of a query. A query is an instruction or process for searching and extracting information from a database. The query can also be used to dictate the manner in which the extracted information is presented. There are various types of queries, and each can be presented in different ways, depending on the specific database system being used. One popular query type is a Boolean query. Such a query in presented in the form of terms and operators. A term corresponds to required information, while the operators indicate a logical relationship between, for example, different terms. There are certain query types that can be presented only in the form of terms. The system receiving the query is then responsible for performing advanced analysis to determine the most appropriate relationships for the terms.
There are various systems that exist for analyzing information. Such analysis can include searching, clustering, and classification. For example, there are a number of systems that allow a query for a search to be received as input in order to retrieve a set of documents from a database. There are other systems that will take a set of documents and cluster them together based on prescribed criteria. There are systems that, given a set of topics or categories, will receive and assign new documents to one of those categories.
As used herein, clustering can be defined as a process of grouping items into different unspecified categories based on certain features of the items. In the case of document clustering, this can be considered as the grouping of documents into different categories based on topic (i.e., literature, physics, chemistry, etc.). Alternatively, the collection of items can be provided in conjunction with some fixed number of pre-defined categories or bins. The items would then be classified or assigned to the respective bins, and the process is referred to as classification.
Most current systems perform search, clustering, and classification based on key words or other syntactic (i.e., word-based) level of analysis of the documents. These systems have the disadvantage that their performance is restricted by their ability to match only on the level of individual words. For example, such systems are unable to decipher whether a particular word is used in a different context within different documents. Further, such systems are unable to recognize when two different words have substantially identical meanings (i.e., mean the same thing). Consequently, the results of a search will often contain irrelevant documents. Such systems are also highly dependent on a user's knowledge of a subject area for selecting terms that most accurately represent the desired results. Another disadvantage of current systems is the inability to accurately cluster and classify documents. This inability is due, in part, because of the restriction to matching on the level of individual words.
Consequently, such systems are unable to accurately perform high level searching, clustering, and classification. Such systems are also often unable to perform these tasks with a high degree of efficiency, especially when documents can be hundreds or thousands of pages long and when vocabularies can cover millions of words.
Accordingly, there exists a need for representing information at a level that does not restrict searching to the level of individual words. There also exists a need for automatically training this semantic representation to allow customized representations in different domains. There also exists a need for an ability to cluster and classify information based on a higher level than individual words.
SUMMARY OF THE INVENTION
An advantage of the present invention is the ability to represent information on a semantic level. Another advantage of the present advantage is the ability to automatically customize the semantic level based on user-defined topics. Another advantage is the ability to automatically train new semantic representations based solely on sample assignments to categories. A further advantage of this invention is the ability to automatically create a semantic lexicon, rather than requiring that a pre-constructed lexicon be supplied. A further advantage is the ability to construct semantic representations without the need to perform difficult and expensive linguistic tasks such as deep parsing and full word-sense disambiguation. A still further advantage is the ability to scale to real-world problems involving hundreds of thousands of terms, millions of documents, and thousands of categories. A still further advantage of the present invention is the ability to search, clusters, and classify information based on its semantic representation.
These and other advantages are achieved by the present invention wherein a trainable semantic vector (TSV) is used to provide a semantic representation of information or items, such as documents, in order facilitate operations such as searching, clustering, and classification on a semantic level.
According to one aspect of the invention, a method of constructing a TSV representative of a data point in a semantic space comprises the steps: constructing a table for storing information indicative of a relationship between predetermined data points and predetermined categories corresponding to dimensions in a multi-dimensional semantic space; determining the significance of a selected data point with respect to each of the predetermined categories; constructing a trainable semantic vector for the selected data point, wherein the trainable semantic vector has dimensions equal to the number of predetermined categories and represents the strength of the data point with respect to the predetermined categories. The data point can correspond to various types of information such as, for example, words, phrases, sentences, colors, typography, punctuation, pictures, arbitrary character strings, etc. The TSV results in a representation of the data point at a higher (i.e., semantic) level.
According to another aspect of the invention, a method of producing a semantic representation of a dataset in a semantic space comprises the steps: constructing a table for storing information indicative of a relationship between predetermined data points within the dataset and predetermined categories corresponding to dimensions in a multi-dimensional semantic space; determining the significance of each data point with respect to the predetermined categories; constructing a trainable semantic vector for each data point, wherein each trainable semantic vector has dimensions equal to the number of predetermined categories and represents the relative strength of its corresponding data point with respect to each of the predetermined categories; and combining the trainable semantic vectors for the data points in the dataset to form the semantic representation of the dataset. Such a method advantageously allows both datasets and the data po
Calistri-Yeh Randall J.
Osborne George B.
Snyder David L.
Yuan Bo
Manning & Napier Information Services, LLC.
McDermott & Will & Emery
Pannala Sathyanaraya R
Robinson Greta
LandOfFree
Construction of trainable semantic vectors and clustering,... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Construction of trainable semantic vectors and clustering,..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Construction of trainable semantic vectors and clustering,... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3308327