Data processing: speech signal processing – linguistics – language – Linguistics
Reexamination Certificate
1999-05-05
2002-09-03
Edouard, Patrick N. (Department: 2644)
Data processing: speech signal processing, linguistics, language
Linguistics
C704S009000
Reexamination Certificate
active
06446035
ABSTRACT:
FIELD OF THE INVENTION
The invention relates to techniques that find groups of people based on behavior.
BACKGROUND
Various conventional techniques have been developed to find groups of people based on behavior. Well-known examples include techniques for creating mailing lists or phone lists based on behavior such as membership in an organization, occupation, or product purchasing behavior, and so forth. Such techniques are frequently employed to target marketing activities, such as mailed advertisements or telemarketing.
Techniques have also been proposed for obtaining information about browsing behavior on the World Wide Web (“WWW” or “the Web”).
ISYS HindSite, a product of ISYS/Odyssey Development Inc., described at http://www.isysdev.com/products/hindsite.htm, saves information about where a Web user has been and what the user has seen. The user can perform full text searches on the contents of previously accessed Web pages, even when bookmarks have not been created. Although Netscape Navigator's history facility lists the universal resource locations (URLs) visited in a Web session, HindSite can index every word of every Web page accessed over a timeframe from one week to six months. HindSite's Plain English query allows users to quickly search by making a statement or asking a question in plain English.
Pirolli, P., Pitkow, J., and Rao, R., “Silk from a Sow's Ear: Extracting Usable Structures from the Web”,
Conference on Human Factors in Computing Systems
(CHI 96), Vancouver, B.C., Canada, Apr. 13-18 1996, describe techniques that utilize topology and textual similarity between items as well as usage data collected by servers and page meta-information like title and size to form document collections. Pages can be related because they have been collected by a particular community or organization. Categorization and associative retrieval techniques provide a means for monitoring the interaction of users and WWW pages. Data extracted from access logs can include topology, page meta-information, usage frequency and usage paths, and text similarity among all text WWW pages at a Web locality. Servers have the ability to record transactional information consisting of at least the time, the name of the URL being requested, and the machine name making the request. When multiple users from a machine name are suspected, heuristics can be used to disambiguate user paths.
Pirolli et al. also describe techniques that tokenize the text for each WWW page and index the tokenized text using a full-text retrieval engine. Document vectors for a pair of pages can be used to obtain a similarity measure between the two pages. Activation network techniques can be applied to the extracted data for purposes such as predicting the interests of home page visitors or assessing the typical web author at a locality.
SUMMARY OF THE INVENTION
The invention addresses problems that arise in finding groups of people. It is often useful to act in relation to a group of people rather than in relation to an entire population that includes the group. For example, it is often much more efficient to target an advertisement or other message to a group of people who are likely to be interested rather than to the entire population. Similarly, if one is searching for people who meet a description, it can be much more efficient to search over a relatively small group of people likely to meet the description than to search the entire population. Acting in relation to a smaller group rather than an entire population can be beneficial even with smaller populations, such as a company, a workgroup, or a community.
Conventional mailing list techniques, mentioned above, typically depend on relatively superficial information about people, such as occupation, membership in organizations, product purchasing behavior, and the like. As a result, the conventional techniques may not discover groupings of people based on more subtle facts about their behavior.
In general, conventional mailing list techniques also neglect sources of information that have recently become available due to technological advances. For example, many systems have been developed in recent years to provide access to resources such as documents in electronic form. The World Wide Web (“WWW” or the “Web”) is an example of such a system that has come into widespread use. Other systems that provide access to resources in electronic form include computers and other devices that can be used to access documents and other resources, and scanners, printers, and digital copiers, in which a resource may be accessed to create an electronic version or for the purpose of providing an electronic version in a print or copy job.
Conversely, conventional techniques for gathering information about resource access behavior do not provide information about groups within a population. For example, HindSite, described above, gathers information about one person's browsing history. But information about one person obviously does not provide information about groups of people. Therefore, HindSite could not provide information about groups.
Other conventional techniques, exemplified by the above-described Pirolli et al. article, are designed to gather and analyze information about browsing behavior of large numbers of users in a relatively anonymous manner. Although such information can be highly informative, these techniques have not been applied to the problems of grouping people.
The invention alleviates these problems by providing techniques that can find groups of people using information about resources the people have accessed. The techniques are applicable where the accessed resources include linguistically analyzable content, such as data defining text or speech. The techniques obtain expression/person data that identify, for each of a set of expression types that occur in the content of the resources, at least one person in the population who has accessed a resource that includes an instance of that type. The techniques use the expression/person data to obtain group information that can indicate a group of people in the population who have accessed resources that include instances of expression types that have similar conceptual content.
The new techniques can be implemented in a system in which resources can be accessed through a network, such as a system that accesses Web pages through the Internet or an intranet. The linguistically analyzable content can be text. For example, text in an accessed Web page can be used to obtain an item of type data indicating an expression type that occurs in the text, such as by performing linguistic analysis. The item of type data can then be associated with an identifier of the person who accessed the Web page, such as a logon name, to obtain an item of expression/person data.
The expression/person data can be stored in a database and the group information can be obtained in response to a query signal from a user. For example, the query signal can indicate a set of expressions, such as a set of words relating to a topic. The query signal can be used to access the expression/person data and obtain output data indicating a group of people who have accessed resources that include expressions having similar conceptual content. Information about the indicated group can then be presented to the user. As a result, the user can find a group of people likely to be interested in the same topic.
Group information could alternatively be obtained by comparing personal profiles. For example, the profile for each person could indicate expression types occurring in resources the person has accessed. Two personal profiles could be compared to find pairs of expressions that have similar conceptual content, with the number of such pairs being a measure of similarity between two people's behavior.
The expression/person data can also indicate resource handles, such as universal resource locations (URLs), that can be used to access resources that include instances of an expression type. The resource handles can be presented togeth
Grefenstette Gregory
Roux Claude
Edouard Patrick N.
Xerox Corporation
LandOfFree
Finding groups of people based on linguistically analyzable... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Finding groups of people based on linguistically analyzable..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Finding groups of people based on linguistically analyzable... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2822257