Data processing: database and file management or data structures – Database design – Data structure types
Reexamination Certificate
2001-08-10
2004-05-04
Mizrahi, Diane D. (Department: 2175)
Data processing: database and file management or data structures
Database design
Data structure types
C707S793000
Reexamination Certificate
active
06732097
ABSTRACT:
BACKGROUND OF THE INVENTION
The inventions herein relate to systems and methods for desired information located within one or more text documents. More particularly, the inventions relate to systems and methods which permit rapid, resource-efficient searches of natural language documents in order to locate pertinent documents and passages based on the role(s) of the user's search term.
In order to facilitate discussion of the prior art and the inventions with precision, the terms below are defined for the reader's convenience.
Glossary
Information Retrieval (IR)—The task of searching for textual information that matches a user's query from a set of documents.
Information Extraction (IE)—The task of identifying very specific elements, defined by a user, in a text. Often, this is the process of answering the questions who, what, where, when, how, and why. For example, a user might be interested in extracting the names of companies that produce software and the names of those software packages. Information Extraction is distinct from Information Retrieval because 1) IE looks for specific information within a document rather than returning an entire document, and 2) an IE system is preprogrammed for these specifications while an IR system must be general enough to respond to any user query.
Relevance—A document is relevant if it matches the user's query.
Recall—A measure of performance. Given the total number of documents relevant to a user's query, recall is the percentage of that number that the system returned as relevant. For example, if there are 500 documents that match a user's query, but the IR system only returns 50 relevant documents, then the system has demonstrated 10% recall.
Precision—A measure of performance. Given the total number of documents truly relevant to a user's query, precision is the percentage of the returned documents that were truly relevant. For example, if the IR system returned 50 documents, but only 25 of them matched the query, the system has demonstrated 50% precision.
Syntactic Roles—The subject, direct object, and indirect object of a clause. Although not strictly a syntactic role, we also include the type of verb phrase (active-voice, passive-voiced, middle-voiced, infinitive) in this group.
Conceptual Roles—Conceptual roles are a way of identifying the particular players within an action or event without regard to the syntax of the clause in which the action or event occurs. Consider the following two sentences.
1. The boy purchased an ice cream cone.
2. An ice cream cone was purchased by the boy.
In the first sentence, the subject is the purchaser and the direct object is the item that was purchased. In the second sentence, however, the subject is now the thing that was purchased and the purchaser is the object of the prepositional phrase introduced by “by.” The “purchaser” and “purchased object” represent conceptual roles because they correspond to specific participants in a purchasing event. As evidenced by these two sentences, conceptual roles can appear in different locations within a sentence's syntactic structure. The advantage of using conceptual roles for information extraction over syntactic roles is that a system can extract the participants of an event regardless of the particular syntax of the sentence.
Theta Roles—Theta roles (also called thematic roles) are similar to conceptual roles in that they correspond to the participants of events or actions. In contrast to conceptual roles, the set of theta roles as defined herein is relatively constrained to include actors (who perform actions), objects or recipients (who receive action), experiencers (actors which play a role but receive no action directly), instruments (used to perform an action), dates (when an action occurred) and locations (where an action occurred). The set of conceptual roles, however, is not constrained. Conceptual roles can be defined to be appropriate to a particular task or collection of texts. In terrorism texts, for example, we may want to define the conceptual roles of perpetrator and victim, while in corporate acquisition texts we may want to define the conceptual roles of purchaser, purchasee, and transaction amount.
Syntactic Caseframe—An extraction pattern based purely on syntactic roles, e.g. “SUBJ <active-voice:kidnap>” would extract the subject of any active-voice construction of the verb “to kidnap.”
Caseframe—synonymous with syntactic caseframe.
Theta Caseframe—A caseframe based on theta roles (often called conceptual roles) rather than syntactic roles, e.g. “AGENT <verb:purchase>” or “OBJECT <verb:purchase>.”
Morphological Root Form—The original form of a word once suffixes and prefixes have been removed, e.g. verb conjugations reduced to the raw verb form: “reported” and “reporting” are both forms of “report.”
Associative Model—The traditional approach to recognizing meaning in text. This model recognizes that certain words in association with each other generate meaning. For example, the terms “headquarters,” “smoke,” “alarm” and “siren” appear to generate the concept of a headquarters building on fire even though the term “fire” does not occur. Compare this approach to the Relational Model below.
Relational Model—An approach to recognizing meaning in text that takes advantage of the relationships between words. For example, the following three phrases each generate a different meaning: “headquarters on fire,” “headquarters under fire” and “fire headquarters.” The key to recognizing the distinction among these phrases is to recognize the relationship between “headquarters” and “fire.”
Relational Text Index (RTI)—The final output which may be generated when using the invention. This is an index of events, relationships, the participants in those events or relationships, along with which document and sentence they occurred in.
Meta-type: A way of collecting specific conceptual types into a more general type. For example, if a verb normally represents a particular action, then a meta-type can be a group of verbs that could be considered synonymous. For example, the verbs “to think,” “to believe,” “to understand” could be considered to be somewhat synonymous, and as verbs of cognition, they give rise to the meta-type “Cognitive-action.” Meta-types do not necessarily imply a two-level classification scheme. More than one meta-type may be combined into a single, more general meta-type. The meta-type, “movement-action” contains the meta-types “transportation-action” and “physical-movement-action” in which the former includes “to fly” and “to drive” while the latter includes “to walk,” “to run” and “to crawl.” Meta-types, therefore, represent nodes in a hierarchy of semantically related words in which each meta-type node must have at least two children. Note that common examples of non verb-based meta-types include grouping semantically related nouns or noun phrases together to include collections of dates, times, and locations.
Morphological Root Form—The original form of a word once suffixes and prefixes have been removed, e.g. verb conjugations reduced to the raw verb form: “reported” and “reporting” are both forms of “report.”
POWERDRILL—A particular system that implements some of the inventions herein for information retrieval.
With the terms defined in the glossary above in mind, a discussion of the typical prior art keyword-based information retrieval systems and their weaknesses will be more meaningful.
DISCUSSION OF PRIOR ART
Traditional methods for information retrieval are based on an associative model of recognizing meaning in text. Associative models identify concepts by measuring how often particular terms occur in a specific document compared to how often they occur in general. In practice, this typically means means that such systems record the content of a document by recognizing which words appear within the document along with their frequency. Essentially, a standard information retrieval system will count how often each English word occurs in a particular document. This inf
Bean David L.
Lorenzen Jeffrey A.
Wakefield Todd D.
Attensity Corporation
McCarthy Daniel P.
Parsons Behle & Latimer
Wu Yicun
LandOfFree
Relational text index creation and searching does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Relational text index creation and searching, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Relational text index creation and searching will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3217875