WO2010014082A1 - Procédé et appareil pour associer des ensembles de données à l’aide de vecteurs sémantiques et d'analyses de mots-clés - Google Patents

Procédé et appareil pour associer des ensembles de données à l’aide de vecteurs sémantiques et d'analyses de mots-clés Download PDF

Info

Publication number
WO2010014082A1
WO2010014082A1 PCT/US2008/071505 US2008071505W WO2010014082A1 WO 2010014082 A1 WO2010014082 A1 WO 2010014082A1 US 2008071505 W US2008071505 W US 2008071505W WO 2010014082 A1 WO2010014082 A1 WO 2010014082A1
Authority
WO
WIPO (PCT)
Prior art keywords
dataset
group
subject
keyword
semantic
Prior art date
Application number
PCT/US2008/071505
Other languages
English (en)
Inventor
Wen Ruan
Clint Prentiss Mah
Gerald Francis Healey Iii
Andrew Lawrence Farris
Gabriel Steinberg
Original Assignee
Textwise Llc
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Textwise Llc filed Critical Textwise Llc
Priority to JP2011521074A priority Critical patent/JP2011529600A/ja
Priority to EP08782506A priority patent/EP2307951A4/fr
Priority to PCT/US2008/071505 priority patent/WO2010014082A1/fr
Priority to CN200880001312A priority patent/CN101802776A/zh
Publication of WO2010014082A1 publication Critical patent/WO2010014082A1/fr

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
    • G06F16/33—Querying
    • G06F16/3331—Query processing
    • G06F16/334—Query execution

Definitions

  • This disclosure relates to method and system for identifying contextually related datasets, such as documents, web pages, e-mails, search queries, advertisements, etc., and more specifically, to method and system for identifying datasets that are contextually related to a subject dataset by analyzing unique semantic vectors of the datasets and keyword semantic representations including information of representative keywords of the datasets.
  • Search engines or advertisement placement systems such as those developed by Microsoft Corporation, Google Inc., Vibrant Media or Yahoo! Inc., are widely used to identify documents or files that are potentially relevant to search queries input by users, or to select and display advertisements that are contextually related to one or more datasets, such as documents, e-mail messages, RSS feeds, or web pages, that have been or are being viewed or manipulated by a user.
  • This disclosure describes various embodiments that efficiently identify one or more datasets, such as documents, web pages, e-mails, etc., that may contextually relate to a subject dataset, such as a search query or a web page being viewed by a user, by analyzing unique semantic vectors representing the datasets and semantic representations including information of representative keywords of the datasets.
  • datasets such as documents, web pages, e-mails, etc.
  • An exemplary method controls a data processing system for relating at least one dataset from a group of datasets to a subject dataset.
  • Each dataset or the subject dataset includes at least one keyword.
  • the method accesses a semantic vector representing the subject dataset and a respective semantic vector representing each respective dataset in the group.
  • Each semantic vector representing each respective dataset in the group includes collective information of relationships between each of the at least one keyword in the respective dataset and predetermined categories to which each of the at least one keyword in the respective dataset may relate.
  • the semantic vector representing the subject dataset includes collective information of relationships between each of the at least one keyword in the subject dataset and predetermined categories to which each of the at least one keyword in the subject dataset may relate, and the semantic vector representing the subject dataset or each respective dataset in the group has a dimension equal to the number of the predetermined categories. For each dataset in the group, determining a first similarity between the subject dataset and each dataset in the group by comparing the semantic vector associated with the subject dataset to the semantic vector associated with each dataset in the group. The exemplary method further accesses a keyword semantic representation of the subject dataset and a keyword semantic representation of each respective dataset in the group.
  • the keyword semantic representation of the subject dataset or the keyword semantic representation of each respective dataset in the group includes Information indicative of representative keyword of the subject dataset or the respective dataset in the group, and the keyword semantic representation of the subject dataset or the keyword semantic representation of each respective dataset in the group is constructed in a manner different from the semantic vector of the subject dataset or the semantic vector of each respective dataset in the group. For each dataset in the group, determining a second similarity between the subject dataset and each dataset in the group by comparing the keyword semantic representation of the subject dataset and the keyword semantic representation of each dataset in the group. At least one of the datasets in the group is selected according to the first similarity between the subject dataset and each dataset in the group, and the second similarity between the subject dataset and each dataset in the group.
  • the method relates the at least one selected dataset in the group to the subject dataset.
  • the at least one of the datasets may be presented to a user concurrently with the subject dataset or subsequent to presenting the subject dataset to the user.
  • the at least one of the datasets or the subject dataset may be presented to the user in an audio form, a visual form, a video form, a haptic form, or any combination thereof.
  • At least one of the datasets in the group is an advertisement
  • the subject dataset is a document, a web page, an e-mail, a RSS news feed, a data stream, broadcast data or information related to a user; or a portion or a combination of one or more documents, web pages, e-mails, RSS news feeds, data streams, broadcast data or information related to a user.
  • the exemplary method conveys the at least one selected dataset or a file associated with the selected dataset with the subject dataset or a file associated with the subject dataset, to a user.
  • the at least one selected dataset may be conveyed to the user by displaying the at least one selected dataset, playing an audible signal according to the at least one selected dataset or providing a link to the at least one selected dataset.
  • the at least one keyword includes at least one of a word, a phrase, a character string, a pre-assigned keyword, a sub-dataset, meta information and information retrieved based on a link included in the respective dataset.
  • the semantic vector for each dataset is pre-calculated and included in the respective dataset. The semantic vector may be dynamically generated on the fly.
  • the semantic vector representing each respective dataset in the group is constructed based on at least one keyword of each respective dataset in the group and known relationships between known keywords and predetermined categories to which the known keywords may relate
  • the semantic vector representing the subject dataset is constructed based on at least one keyword of the subject dataset and the known relationships between known keywords and predetermined categories to which the known keywords may relate.
  • the semantic vector associated with the respective dataset is generated further based on information related to at least one user or at least one dataset linked to the respective dataset.
  • the information related to the at least one user may include at least one of documents previously viewed, previous search requests, user preferences and personal information.
  • the step of selecting at least one of the datasets in the group according to the first similarity between the subject dataset and each dataset in the group, and the second similarity between the subject dataset and each dataset in the group comprising designating one of the first similarity and the second similarity as a primary similarity and the other as a secondary similarity, accessing information of a plurality of preset relevance levels for the primary similarity; for each dataset in the group, mapping the primary similarity to one of the preset relevance levels according to the primary similarity; ranking the datasets in the group according to respective mapped preset relevance levels of the datasets in the group; within each relevance level, ranking the datasets in each relevance level according to the secondary similarity of the datasets; and selecting the at least one of the datasets in the group according to a result of ranking the datasets in each relevance level.
  • the step of selecting at least one of the datasets in the group according to the first similarity between the subject dataset and each dataset in the group, and the second similarity between the subject dataset and each dataset in the group comprising: designating one of the first similarity and the second similarity as a primary similarity and the other as a secondary similarity; ranking the datasets in the group according to the primary similarity; selecting at least one candidate dataset from the ranked datasets according to a preset criteria; ranking the at least one candidate dataset according to the secondary similarity; selecting the at least one of the datasets in the group according to a result of ranking the at least one candidate dataset.
  • the step of selecting at least one of the datasets in the group according to the first similarity between the subject dataset and each dataset in the group, and the second similarity between the subject dataset and each dataset in the group comprising: for each dataset in the group, calculating a composite similarity based on a respective first similarity of the dataset and a respective second similarity of the dataset according to a preset formula; selecting the at least one of the datasets in the group according to respective composite similarities of the datasets.
  • An exemplary data processing system for relating at least one dataset from a group of datasets to a subject dataset.
  • Each dataset or the subject dataset includes at least one keyword.
  • the system includes a data processor configured to process data and a data storage system configured to store instructions which, upon execution by the data processor, control the data processor to perform prescribed steps.
  • the steps include accessing a semantic vector representing the subject dataset and a respective semantic vector representing each respective dataset in the group, wherein: each semantic vector representing each respective dataset in the group includes collective information of relationships between each of the at least one keyword in the respective dataset and predetermined categories to which each of the at least one keyword in the respective dataset may relate, the semantic vector representing the subject dataset includes collective information of relationships between each of the at least one keyword in the subject dataset and predetermined categories to which each of the at least one keyword in the subject dataset may relate, and the semantic vector representing the subject dataset or each respective dataset in the group has a dimension equal to the number of the predetermined categories; for each dataset in the group, determining a first similarity between the subject dataset and each dataset in the group by comparing the semantic vector associated with the subject dataset to the semantic vector associated with each dataset in the group; accessing a keyword semantic representation of the subject dataset and a keyword semantic representation of each respective dataset in the group, wherein: the keyword semantic representation of the subject dataset or the keyword semantic representation of each respective dataset in the group includes information indicative of representative keyword of the subject
  • An embodiment of this disclosure includes a machine-readable medium carrying instructions which, upon execution of a data processing system, control the data processing system to perform machine-implemented steps to relate at least one dataset from a group of datasets to a subject dataset. Each dataset or the subject dataset includes at least one keyword.
  • the steps comprises accessing a semantic vector representing the subject dataset and a respective semantic vector representing each respective dataset in the group, wherein: each semantic vector representing each respective dataset in the group includes collective information of relationships between each of the at least one keyword in the respective dataset and predetermined categories to which each of the at least one keyword in the respective dataset may relate, the semantic vector representing the subject dataset includes collective information of relationships between each of the at least one keyword in the subject dataset and predetermined categories to which each of the at least one keyword in the subject dataset may relate, and the semantic vector representing the subject dataset or each respective dataset in the group has a dimension equal to the number of the predetermined categories; for each dataset in the group, determining a first similarity between the subject dataset and each dataset in the group by comparing the semantic vector associated with the subject dataset to the semantic vector associated with each dataset in the group; accessing a keyword semantic representation of the subject dataset and a keyword semantic representation of each respective dataset in the group, wherein: the keyword semantic representation of the subject dataset or the keyword semantic representation of each respective dataset in the group includes information indicative of representative keyword of the subject
  • Figure 1 is a block diagram of an exemplary advertisement placement system
  • Figure 2 shows an embodiment of an exemplary advertisement placement system according to this disclosure
  • Figure 3 illustrates the operation of another embodiment of an advertisement placement system according to this disclosure.
  • Figure 4 is an exemplary table showing relationships between words and categories
  • Figure 5 is an exemplary table illustrating values corresponding to the significance of the words from Figure 4.
  • Figure 6 is an exemplary table illustrating a representation of the words from Figure 4 in a semantic space.
  • Figure 7 is a block diagram of an exemplary computer system upon which an exemplary advertisement placement system may be implemented.
  • dataset refers to a collection of expressions that are readable and/or understandable by humans and/or machines
  • key refers to one or more elements, such as textual or symbolic elements, numbers, etc., of a dataset.
  • a keyword may be one or more words, phrases, punctuations, symbols and/or sentences contained in the document.
  • a dataset can be a collection of a plurality of different types of datasets, or a portion of a larger dataset.
  • a dataset may be a summary and/or tag summarizing or describing the contents of another dataset. Keywords may or may not be directly viewable to a user. For instance, a keyword may be part of closed captions or hidden subtitles of a video file, lyrics of an audio file, or an element of metadata associated with a Word document. Additional processing may be performed before a keyword can be ascertained or processed by humans or machines. For instance, optical character recognition or voice recognition may be used to convert certain elements in a first format into second format, for easier processing and/or recognition by humans or machines.
  • Examples of datasets include web pages, video, audio or multimedia files, advertisements, e-mails, documents, RSS feeds, multimedia files, photos, figures, drawings, electronic computer documents, sound recordings, broadcasts, video files, metadata, etc., or a collection of one or more of the above.
  • Examples of keywords include words, phrases, symbols, terms, hyperlinks, metadata information, and/or any displayed or un-displayed item(s) included in or associated with a dataset.
  • web pages are understood to refer to any compilation or collection of information that can be displayed in a web browser such as Microsoft Internet Explorer, the content of which may include, but does not limit to, HTML pages, JavaScript pages, XML pages, email messages, and RSS news feeds.
  • subject dataset refers to one or more datasets for which an exemplary system intends to identify one or more datasets, from a group of datasets, that are contextually related to the subject dataset.
  • a subject dataset may be search queries that a user inputs intending to find documents relevant to the search queries; or one or more web pages that an exemplary system according to this disclosure intends to find suitable advertisements for displaying with the web pages.
  • the following examples describe operations of embodiments that identify one or more datasets, such as advertisements, that are contextually related to a subject dataset, such as a web page being reviewed by a user, based on analyses of unique semantic vectors, such as trainable semantic vectors (TSV), that represent the web page and the advertisements, and semantic representations including information of representative keywords of the advertisements and the web page.
  • TSV trainable semantic vectors
  • Various formulas and statistical manipulations can be performed to identify important or representative keywords so that they can be weighed more heavily than others.
  • a trainable semantic vector is a unique type of semantic representations of a dataset and is generated based on data points included in the dataset and known relationships between known data points and predetermined categories. Details of constructions and characteristics of trainable semantic vectors are described in U.S. Patent No. 6,751,621, filed on May 2, 2000 and entitled "CONSTRUCTION OF TRAINABLE SEMANTIC VECTORS AND CLUSTERING," and U.S.
  • Patent Application Serial 11/126,184 (attorney docket No. 55653-019), filed May 11, 2005 and entitled ADVERTISEMENT PLACEMENT METHOD AND SYSTEM USING SEMANTIC ANALYSIS 3 the disclosures of which are incorporated herein by reference in their entireties.
  • FIG. 1 is a diagram of an exemplary advertisement placement system 10 configured to identify, from a group of advertisements 12, one or more advertisements that are contextually related to a web page 11 being viewed by a user, based on analyses of at least two types of semantic representations of advertisements 12 and web page 11 : TSVs and semantic representations including information of representative keywords of advertisements 12 and web page 11.
  • Advertisements 12 may consist of any combination of media, such as text, sound or animation, etc.
  • system 10 Based on results of the analyses, system 10 generates a match result identifying selected advertisements that are contextually related to webpage 12.
  • the selection of one or more advertisements for a particular dataset or web page can occur at the time the dataset is presented, or after or prior to the dataset is presented to a user.
  • advertisement placement system 10 is used to select one or more advertisements 12 that are contextually relevant to webpage 11 such that the webpage is displayed with or linked to the one or more selected advertisements.
  • Datasets that are identified as relevant to a subject dataset are conveyed or presented to a user together with the subject dataset and at different times from the presentation or conveyance of the subject dataset.
  • the datasets may be conveyed or presented to a user in various forms or format, such as audio form, video form, visual form, haptic form, machine-readable format, or any combinations thereof, etc.
  • each of advertisements 12 or web page 11 may be pre-calculated or calculated on the fly.
  • each web page or advertisement includes embedded or associated information of their respective pre- calculated TSVs.
  • the TSV associated with web page 11 is dynamically calculated by system 10.
  • FIG. 2 is a detailed block diagram of an embodiment of advertisement placement system 10.
  • advertisement placement system 10 includes term extractors 102, 112 for identifying and retrieving keywords from advertisements 12 or web page 11.
  • Term extractors 102, 112 perform linguistic analyses on the contents of advertisements 12 or web page 11, to break sentences from advertisements 12 or web page 11 into smaller units, such as words, phrases, etc. Frequently used terms, such as grammatical words like "the", "a”, and so forth, may be removed using a preset stop list. If advertisements 12 or web page 11 includes information other than the actual content (for example, HTML markup tags or JavaScripting), that information may be removed. Software for implementing term extractions is widely available and known to people skilled in the art.
  • Advertisement placement system 10 further includes TSV generators
  • Advertisement placement system 10 includes a TSV indexer 114 and a TSV index database 118, which are used to organize and store generated TSVs for efficient searches.
  • the TSV indexer 114 may be implemented using a full database management system (DBMS) or just a software package for large-scale data record management, and TSV index database 118 may be implemented with a database storing TSV index files including TSVs of advertisements 12 along with links to them. Different indexing schemes may be applied to speed up searching. For example, one common indexing scheme for TSVs is to list them under the individual semantic categories that they reference.
  • TSV matcher 104 determines respective TSV similarities between web page 11 and each advertisement.
  • the similarities may be in the form of a relevance score.
  • the similarity or relevance between TSVs is determined based on a distance between the semantic vectors (TSVs), such as determining N-dimensional Euclidean distance between the TSVs, where N is the number of dimensions of the semantic space or the predetermined categories.
  • TSVs semantic vectors
  • Other comparison methods such as cosine measure, Hamming distance, Minkowski distance or Mahalanobis distance can also be used.
  • Various optimizations can be performed to improve the comparison time including reducing the dimensionality of the TSVs prior to comparison and applying filters to eliminate certain advertisements prior to or subsequent to comparison.
  • the TSV matcher 104 Based on TSV comparison results, the TSV matcher 104 generates a
  • TSV match list 105 including a ranked list of matched advertisements selected from advertisements 12, according to their respective TSV similarities to web page 11.
  • a preset threshold may be applied to select only those advertisements having a degree of similarity beyond a preset threshold.
  • Advertisement placement system 10 further includes mechanism for determining and comparing contextual representations, having a type different from TSVs, for web page 11 and advertisements 12.
  • advertisement placement system 10 generates semantic representations including information of representative keywords of web page 11 and advertisements 12.
  • keyword selectors 115, 106 input terms retrieved by term extractors 102, 112, and select a subset of keywords from the contents of web page 11 or advertisements 12 for representing web page 11 or each of advertisements 12, according to one or more metrics, such as term frequency (how often a term occurs in the page), inverse document frequency (what fraction of pages in a collection include the term), or other approaches well-known to people skilled in the art.
  • keyword selectors 115, 106 may calculate the frequency or the number of appearance of each text in web page 11 or each advertisement, and select representative keywords based on the calculated frequency or the numbers of appearance of each text.
  • Another example is to use stop lists to remove keywords that provide little information about the subject of web page 11 or advertisements 12.
  • Term extractors 102, 112 maintain, or have access to, a stop list including the most commonly occurring words that provide little information about the subject. Keywords included in the stop list are not good search terms.
  • the stop list may be created by a linguistic expert, by an automatic analysis (such as statistical), or by a user or by a combination of all three. It is understood that other approaches known to people skilled in the art may be used to select keywords from web page 11 or advertisements 12 for representing web page 11 or advertisements 12.
  • a keyword index database 117 is provided to store the representative keywords and links to respective advertisements 12.
  • a keyword matcher 107 is provided to determine a keyword similarity between web page 11 and each of advertisements 12, based on information of selected keywords representing each respective advertisement and web page 11.
  • the keyword matcher 107 looks up the set of selected keywords for web page 11 in the keyword index database 117, and generates a keyword relevance score for each advertisement and web page 11, according to one or more known algorithms. For example, a relevance score between two sets of representative keywords is calculated based on the number of matching or common keywords (one term, one vote) included in the advertisement and the web page.
  • the keyword matcher 107 employs more elaborate voting schemes (electoral college, weighted shares, aristocracy with absolute veto, loudness of support) to determine a degree of similarity between each advertisement and web page 11.
  • Other types of calculations such as a vector space model, may use a straight or modified cosine similarity measure to calculate a relevance score.
  • the keyword matcher 107 After the keyword matcher 107 calculates the respective similarities between web page 11 and each respective advertisement, the keyword matcher 107 generates a keyword match list 108 ranking advertisements 12 based on their respective similarities to web page 11 or their respective relevance scores.
  • the TSV match list 105 and the keyword match list 108 are sent to a combiner 109 which generates a final match list 110 according to information included in keyword match list 108 and TSV match list 105.
  • combiner 109 calculates a composite relevance score based on its relevance score in TSV match list 105 and keyword match list 110.
  • a final match list 110 is then generated according to the respective composite relevance scores of advertisements.
  • the composite relevance score is calculated as follows:
  • the coefficients a ls a 2 , b la b 3 , C 1 , C 2 , C 3 may be chosen in a way that equations (2) and (3) are special cases of equation (1).
  • the relevance scores in either or all match lists may be normalized to [0, 1].
  • Conditional or unconditional thresholds may be applied to the relevance scores in either or all match lists to shorten the lists.
  • a final match list 110 is compiled according to the composite scores of the advertisements.
  • advertisements in the TSV match list 105 and keyword match list 108 are rearranged to form an exemplary final match list 110, using a unique formula.
  • Each advertisement in the TSV match list 105 and keyword match list 108 is associated with a respective TSV relevance score and a keyword relevance score.
  • TSV match list 105 ranks advertisements according to their respective TSV relevance scores
  • keyword match list 108 ranks advertisements based on their respective keyword relevance scores.
  • One of TS V relevance score and keyword relevance score is designated as the primary relevance score and the other is designated as the secondary relevance score.
  • Table 1 shows exemplary rank lists having TSV relevance score as the primary relevance score and keyword relevance score as the secondary relevance score.
  • the primary relevance score for each advertisement is mapped into preset relevance levels corresponding to specific ranges of relevance scores.
  • Advertisements are then ranked according to their mapped relevance levels.
  • the secondary relevance score for each respective advertisement is used to rank advertisements within each relevance level.
  • advertisements are re-ranked according to their respective relevance levels. Advertisements within each respective relevance level are then re-ranked according to their respective secondary relevance score. A re-ranked result is shown in Table 2. Column 1 of Table 2 is the final relevance ranking of the advertisements.
  • Advertisement placement system 10 selects one or more advertisements from the final match list 110 for relating to web page 11, according to the ranking of the final match list 110. According to one embodiment, the selected advertisements are displayed with, or linked to, web page 11.
  • system 10 may generate final match list 100 by relying mainly on only one of TSV match list 105 and keyword match list 108. For instance, system 10 relies on keyword match list 108 which selects a preset number of advertisements according to their respective keyword relevance scores. A TSV relevance score for each advertisement is still calculated. Advertisements on keyword rank list 108 are then re-ranked based on their respective TSV relevance scores. System 10 outputs the re-ranked match list as final match list 110.
  • Figure 3 shows another exemplary advertisement placement system
  • TSVs and keyword semantic representations for advertisements 12 are stored within a database 212.
  • database 212 provides two data fields, one for TSV and one for keyword semantic representation.
  • Advertisement placement system 20 further includes a TSV and keyword indexer 211 for organizing and managing TSVs and keyword semantic representations.
  • TSV and keyword indexer 211 may be implemented using a full database management system (DBMS) or just a software package for large-scale data record management, and database 212 may be implemented with a database. Different indexing schemes may be applied to speed up searching.
  • DBMS database management system
  • System 20 includes term extractor 102 and 112, TSV generator 103 and 113, keyword selector 106 and 115, all with the same functionalities as described earlier relative to Figure 2.
  • a TSV and keyword combiner 210 properly associates its TSV and keyword semantic representation with the advertisement.
  • a TSV is generated by TSV generator 103 and keyword semantic representation is generated by keyword selector 106.
  • a TSV and keyword combiner 205 associates or links its TSV and keyword semantic representation with web page 11.
  • Information related to TSVs and keyword semantic representations for web page 11 and advertisements 12 are processed by TSV and keyword matcher 206 which performs functions similar to those of TSV matcher 104 and keyword matcher 107 discussed earlier relative to Figure 2. Relevance scores for TSVs and keyword semantic representations may be calculated in ways similar to those described relative to Figure 2.
  • a final match list 213 is generated by TSV and keyword matcher 206 as discussed earlier with respect to Figure 2.
  • a joint relevance score for each advertisement or each candidate or target dataset may be calculated by combining the keyword semantic representation and the semantic vector representation of a dataset in the same vector space. For instance, both the keyword representation and the semantic vector representation of an advertisement are treated as vectors in the same vector space and combined to form a signal joint semantic vector representation of the advertisement.
  • the semantic vector representation and the keyword semantic representation may be assigned different weightings. For each advertisement, a relevance score is calculated based on the joint semantic vector representation of the advertisement and the joint semantic vector representation of a target dataset.
  • a final match list 213 is generated by the TSV and keyword Matcher 206 according to respective joint relevance scores of the advertisements.
  • match lists generated based on keyword or TSV comparisons can be further refined or re-ranked by other known methods. For instance, datasets or web pages in a rank list may be rearranged using algorithm according to link information between web pages in the final ranking, such as PageRank algorithm developed by Google, Inc., described in U.S. Patent No. 6,285,999, titled "METHOD FOR NODE RANKING IN A LINKED DATABASE," the entire disclosure of which is incorporated herein by reference.
  • TSVs Constructions of TSVs for datasets are now described. Further details of TSVs are described in U.S. Patent No. 6,751,621 and U.S. Patent Application Serial No. 11/126,184, the disclosures of which are previously incorporated by reference.
  • a semantic dictionary is used to find the TSVs corresponding to data points included in the datasets.
  • the semantic dictionary includes known relationships between a plurality of known data points and a plurality of predetermined categories. In other words, the semantic dictionary contains "definitions", i.e., TSVs, of the corresponding words or phrases.
  • An exemplary process for generating a TSV for a dataset using a TSV generator is now described.
  • the dataset can be an advertisement, a web page, or any types of datasets.
  • “words” are used as examples for keywords included in the document. It is understood that many other types of data points or keywords may be included in the document, such as words, phrases, symbols, terms, hyperlinks, metadata information, graphics and/or any displayed or un-displayed item(s) or any combination thereof.
  • the TSV generator Based on input keywords of the document, the TSV generator identifies corresponding keywords in the semantic dictionary and retrieves the respective TSV of each keyword included in the document based on the definitions provided by the semantic dictionary. TSV generator 103 generates the TSV of the document by combining the respective TSVs of the keywords included in the document. For instance, the TSV of the document may be defined as a vector addition of the respective TSVs of all the keywords included in the document.
  • the semantic dictionary is generated by properly determining which predetermined category or categories each of a plurality of known datasets falls into.
  • a sample dataset may fall in more than one predetermined categories, or the sample datasets may be restricted to associate with a single category.
  • a news report related to a patent infringement lawsuit involving a computer company may fall into categories including "intellectual property law", “business controversies”, “operating systems”, “economic issues”, etc., depending on the content of the report and depending on the predetermined categories.
  • the relationships between sample documents and categories can be determined by analyzing the Open Directory Project (ODP), which assigned hundreds of thousands of web pages to a rich topic hierarchy by expert human editors. These sample web pages with assigned categories are called training documents for determining relationships between keywords and predetermined categories. It should be clear to those skilled in the art that other online topic hierarchies, classification schemes, and ontologies can be used in similar ways to relate sample training documents to categories.
  • ODP Open Directory Project
  • ODP category to which the original ODP webpage belongs Optionally filter the web pages to keep only those new web pages that have the same categories as the original
  • ODP web page from which it was derived Remove any web pages that did not download properly, and translate URLs to internal pathnames.
  • ODP categories are removed before processing. These removed categories may include empty categories (categories without corresponding documents), letterbar categories ("movie titles starting with A, B, " with no useful semantic distinction), and other categories that do not contain useful information for identifying semantic content (e.g. empty categories, regional pages in undesired foreign languages) or that contain misleading or inappropriate information (e.g. adult-content pages).
  • ODP category then it is ambiguously classified and may not be a good candidate for training.
  • Optionally adjust the TSV dimensions Inspect the automatically generated TSV dimensions and manually collapse, split, or remove certain dimensions based on the anticipated semantic properties of those dimensions. Types of adjustments could include, but are not limited to, the following. First, if certain words occur frequently in the original category names, those categories can be collapsed to their parent nodes (either because they are all discussing the same thing or because they are not semantically meaningful). Second, certain specific categories can be collapsed to their parents (usually because they are too specific). Third, certain groups of categories separated in the ODP hierarchy can be merged together (for example, "Arts / Magazines and E-Zines / E-Zines" can be merged with "Arts / Online Writing / E-Zines”). [0090] 9.
  • TSV training files For each potential training page, associate that page with the TSV dimension into which the page's category was collapsed. Then select the pages from each TSV dimension that will be used to train that dimension, being careful not to overtrain or undersample. In one embodiment, we randomly select 300 pages that have at least 1000 bytes of converted text (if there are fewer than 300 appropriate pages, we select them all). We then remove any pages longer than 5000 whitespace-delimited words, and we keep a maximum of 200,000 whitespace- delimited words for the entire dimension, starting with the smallest pages and stopping when the cumulative word count reaches 200,000.
  • each dimension starts off with the same label as the ontology path of the ODP category from which it was derived.
  • some labels are manually adjusted to shorten them, make them more readable, and ensure that they reflect the different sub categories that were combined or removed. For example, an original label of "Top / Shopping / Vehicles / motorcycles / Parts_and_Accessories / Harley_Davidson” might be rewritten “Harley Davidson, Parts and Accessories”.
  • the collapse-trim algorithm walks bottom-up through the ODP hierarchy looking at the number of pages available directly in each category node. If there are at least 100 pages stored at that node, then we keep that node as a TSV dimension. Otherwise we collapse it into the parent node. [0093] After the assignment of sample datasets to predetermined categories
  • a data table is created storing information that is indicative of a relationship between keywords included in one or more sample datasets and predetermined categories based on the assignment result.
  • Each entry in the data table establishes a relationship between a keyword and one of the predetermined categories.
  • an entry in the data table can correspond to the number of sample datasets, within a category, that contain a particular keyword.
  • the keywords correspond to the contents of the sample datasets, while the predetermined categories correspond to dimensions of the semantic space.
  • the data table may be used to generate a semantic dictionary that includes "definitions" of each word, phrase, or other keyword within a specific semantic space formed by the predetermined categories, for use in constructing trainable semantic vectors.
  • Figure 4 shows an exemplary data table for constructing a semantic dictionary.
  • table 200 contains rows 410 that correspond to the predetermined categories Cati, Cat2, Cat 3 , Cat 4 , and Cat 5 , and columns 412 representative words W 1 , W 2 , W 3 , W 4 , and W 5 .
  • Each entry 414 within table 200 corresponds to a number of documents that have a particular word, such as one or more of words W 1 , W 2 , W 3 , W 4 , and W 5 , occurring in the corresponding category.
  • word Wi appears a total of 28 times across all categories. In other words, twenty-eight of the documents classified contain word Wi. Examination of an exemplary column 412, such as Cati, reveals that word W 2 appears once in category Cati, word W 3 appears eight times in category Cati, and or W 5 appears twice in category Cati. Word W 4 does not appear at all in category Cati. As previously stated word Wi does not appear in category 1. Referring to row 418, the entry corresponding to category Cat ! indicates that there are eleven documents classified in category Cati. [0098] According to one embodiment, after the data table is constructed, the significance of each entry in the data table is determined.
  • the significance of the entries can, under certain situations, be considered the relative strength with which a word occurs in a particular category, or its relevance to a particular category. Such a relationship, however, should not be considered limiting.
  • the significance of each entry is only restricted to the actual dataset and categories (i.e. features, that are considered significant for representing and describing the category).
  • the significance of each word is determined based on the statistical behavior of the words across all categories. This can be accomplished by first calculating the percentage of keywords occurring in each category according to the following formula:
  • Both u and v represent the strength with which a word is associated with a particular category. For example, if a word occurs in only a small number of datasets from a category but doesn't appear in any other categories, it would have a high v value and a low u value for that category. If the entry appears in most datasets from a category but also appears in several other categories, then it would have a high u value and a low v value for that category.
  • u for each category can be normalized (i.e., divided) by the sum of all values for a keyword, thus allowing an interpretation as a probability distribution.
  • a weighted average of u and v can also be used to determine the significance of keywords, according to the following formula:
  • the variable ⁇ is a weighting factor that can be determined based on the information being represented and analyzed.
  • the weighting factor has a value of about 0.75.
  • Other values can be selected depending on various factors such as the type and quantity of information, or the level of detail necessary to represent the information.
  • Figure 5 illustrates the operation of the above-described manipulating process based on the data from Figure 4.
  • a table 230 stores the values that indicate the relative strength of each word with respect to the categories. Specifically, the percentage of keywords occurring in each category (i.e., u) is presented in the form of a vector for each word. The value for each entry in the u vector is calculated according to the following formula:
  • Table 230 also presents the probability distribution of a keyword' s occurrence across all categories (i.e., v) in the form of a vector for each word.
  • entry) (word n , category m )/word n _totai
  • Table 250 is shown for illustrating the semantic representation or "definition" of the words from Figure 4.
  • Table 250 is a combination of five TSVs that correspond to the semantic representation of each word across the semantic space.
  • the first row corresponds to the TSV of word W 1 .
  • Each TSV has dimensions that correspond to the predetermined categories.
  • the TSVs for words W 1 , W 2 , W 3 , W 4 , and W 5 are calculated according to an embodiment of the disclosure wherein the entries are scaled to optimize the significance of the word with respect to that particular category. More particularly, the following formula is used to calculate the values.
  • the entries for each TSV are calculated based on the actual values stored in table 230. Accordingly, the TSVs shown in table 250 correspond to the "definition" of the exemplary words W 1 , W 2 , W 3 , W 4 , and W 5 represented in Figure 4 relative to each predetermined categories or vector dimension, which collectively compose a semantic dictionary for the semantic space formed by the predetermined categories.
  • TSV The application of TSV is not restricted to just one language. As long as appropriate sample datasets are available, it is possible to build a semantic dictionary for different languages. For instance, English sample datasets from the Open Directory Project can be replaced with suitable sample datasets in another language in generating the semantic dictionary. There can be a separate semantic dictionary for each language. Alternately, the keywords for all languages can reside in a single common semantic dictionary. Different languages may share the same predetermined categories or semantic dimensions, or may have completely different predetermined categories or semantic dimensions, depending on whether they share the same semantic dictionary and whether it is desired to compare semantic vectors across languages. [00110] After the semantic dictionary is created, the semantic dictionary can be accessed by TSV generator 103 to find corresponding TSVs for keywords included in the target document.
  • the TSVs of the keywords included in the target document are combined to generate the TSV of the target document.
  • the manner in which the TSVs are combined depends upon the specific implementations.
  • the TSVs may be combined using a vector addition operation.
  • TSV for a document can be represented as follows:
  • TSV (document) TS V(W1)+TS V(W2)+TS V(W3)...+TSV(WN) where Wl, W2, W3,... WN are words included in the document.
  • the generation of TSVs for datasets may utilize many types information including keywords in the datasets, information retrieved based on keywords included in the advertisements and datasets, and additional information assigned to the datasets.
  • the generation of TSVs for advertisements may be performed based on information including, but not limited to, words displayed in the advertisements, a set of keywords associated with each advertisement, the title of the advertisement, a brief description of the advertisement, marketing literature associated with the advertisement that describes the item being advertised or the audience to which it is being sold, and information from web sites that may be referenced by the advertisement.
  • TSVs for web pages may be performed based on information including, but not limited to, some or all of the actual text that appears on the web page, meta-text fields associated with the web page such as title, keywords, and description, text from other web pages linked to or linked by the web page, etc.
  • the TSVs for advertisements can be generated off-line and updated as advertisements are modified, added, or removed. But TSVs can also optionally be generated at the time of advertisement placement. Similarly, TSVs for web pages or other datasets can be generated either off-line or on the fly.
  • an exemplary system disclosed herein analyzes various sections of a dataset, such as a web page or displayed document, and automatically links each section one or more descriptions to a set of background articles, such as encyclopedic articles from Wikipedia (http://www.wikipedia.org), based on a final match list of the background articles.
  • encyclopedic articles from Wikipedia (http://www.wikipedia.org)
  • the methods and systems disclosed herein are applicable to various purposes, such as associating one or more advertisements to one or more web pages or documents, or vice versa; retrieving related documents based on a user's search queries; finding background information for different portions of a dataset, and the like.
  • a dataset as used herein may include only a single type of dataset, such as web page(s) or document(s), or a collection of different types of datasets, such as a combination of e-mails and web pages, documents and broadcasting data.
  • Another embodiment according to this disclosure utilizes a refined representation called "tagged key" to represent and index datasets, such as advertisements 12 and web page 11.
  • a tagged key associates a keyword found in a dataset with one or more specific semantic categories applicable to the dataset.
  • the term "bank” may carry many different meanings, but when it is tagged with a semantic category such as Financial Institution, one will no longer match it with "bank” tagged with a semantic category such as Geological Structure.
  • candidate keywords that are considered to be representing the web page or advertisement are selected from each advertisement or web page 11 by keyword selector 115 or 106 as discussed earlier relative to Fig. 3.
  • candidate keywords may be selected based on the frequency of each keyword appearing in a specific dataset or document.
  • An exemplary system according to this disclosure accesses a semantic dictionary for information related to predetermined semantic categories and their relationships to the candidate keywords. For instance, for a data set having N candidate keywords and M predetermined categories, MxN pairs of keyword and category (possible tagged keys) are available.
  • a filter may be used to eliminate categories that are less relevant to a keyword.
  • a threshold specifying minimum requirement of relevance may be used to identify categories that are sufficient relevant to the keyword.
  • One exemplary way to select categories for a keyword is simply looking in a semantic dictionary as discussed above, which includes information specifying how strongly a particular term selects for a given semantic category. In one embodiment, most strongly selected category or categories for a keyword would be the primary candidate for tagging.
  • K2 K2.
  • a keyword is related to more than one category, such as categories Cl, C2, C3, and C4, then one has several options: (1) choose the category with the strongest connection to the keyword; (2) choose all the categories with connections above a minimum threshold; or (3) choose all categories regardless of strength of connection.
  • the result will be a list of paired categories and keywords, the tagged keys, such as Kl+Cl, K1+C2, and K2+C4, etc., for representing a data set.
  • FIG. 7 is a block diagram that illustrates a computer system 100 upon which an exemplary system of this disclosure may be implemented.
  • Computer system 100 includes a bus 702 or other communication mechanism for communicating information, and a processor 704 coupled with bus 702 for processing information.
  • Computer system 100 also includes a main memory 706, such as a random access memory (RAM) or other dynamic storage device, coupled to bus 702 for storing information and instructions to be executed by processor 704.
  • main memory 706 such as a random access memory (RAM) or other dynamic storage device
  • Main memory 706 also may be used for storing temporary variables or other intermediate information during execution of instructions to be executed by processor 704.
  • Computer system 100 further includes a read only memory (ROM) 708 or other static storage device coupled to bus 702 for storing static information and instructions for processor 704.
  • ROM read only memory
  • a storage device 710 such as a magnetic disk or optical disk, is provided and coupled to bus 702 for storing information and instructions.
  • Computer system 100 may be coupled via bus 702 to a display 712, such as a cathode ray tube (CRT), for displaying information to a computer user.
  • a display 712 such as a cathode ray tube (CRT)
  • An input device 714 is coupled to bus 702 for communicating information and command selections to processor 704.
  • cursor control 716 is Another type of user input device
  • cursor control 716 such as a mouse, a trackball, or cursor direction keys for communicating direction information and command selections to processor 704 and for controlling cursor movement on display 712.
  • This input device typically has two degrees of freedom in two axes, a first axis (e.g., x) and a second axis (e.g., y), that allows the device to specify positions in a plane.
  • construction of TSVs and semantic operations is provided by computer system 100 in response to processor 704 executing one or more sequences of one or more instructions contained in main memory 706 or storage device 710, or received from the network link 120.
  • Such instructions may be read into main memory 706 from another computer-readable medium, such as storage device 710.
  • Execution of the sequences of instructions contained in main memory 706 causes processor 704 to perform the process steps described herein.
  • processors in a multi-processing arrangement may also be employed to execute the sequences of instructions contained in main memory 706.
  • hard-wired circuitry may be used in place of or in combination with software instructions to implement the disclosure.
  • Non-volatile media include, for example, optical or magnetic disks, such as storage device 710.
  • Volatile media include dynamic memory, such as main memory 706.
  • Transmission media include coaxial cables, copper wire and fiber optics, including the wires that comprise bus 702.
  • Computer-readable media include, for example, a floppy disk, a flexible disk, hard disk, magnetic tape, any other magnetic medium, a CD-ROM, DVD, any other optical medium, punch cards, paper tape, any other physical medium with patterns of holes, a RAM, a PROM, and EPROM, a FLASH-EPROM, any other memory chip or cartridge, or any other medium from which a computer can read.
  • Various forms of computer readable media may be involved in carrying one or more sequences of one or more instructions to processor 704 for execution.
  • the instructions may initially be borne on a magnetic disk of a remote computer.
  • the remote computer can load the instructions into its dynamic memory and send the instructions over a telephone line using a modem.
  • a modem local to computer system 100 can receive the data on the telephone line and use an infrared transmitter to convert the data to an infrared signal.
  • An infrared detector coupled to bus 702 can receive the data carried in the infrared signal and place the data on bus 702.
  • Bus 702 carries the data to main memory 706, from which processor 704 retrieves and executes the instructions.
  • Computer system 100 also includes a communication interface 718 coupled to bus 702.
  • Communication interface 718 provides a two-way data communication coupling to a network link 120 that is connected to a local network 722.
  • communication interface 718 may be an integrated services digital network (ISDN) card or a modem to provide a data communication connection to a corresponding type of telephone line.
  • ISDN integrated services digital network
  • communication interface 718 may be a local area network (LAN) card to provide a data communication connection to a compatible LAN.
  • LAN local area network
  • Wireless links may also be implemented.
  • communication interface 718 sends and receives electrical, electromagnetic or optical signals that carry digital data streams representing various types of information.
  • Network link 120 typically provides data communication through one or more networks to other data devices.
  • network link 120 may provide a connection through local network 722 to a host computer 724 or to data equipment operated by an Internet Service Provider (ISP) 726.
  • ISP 726 in turn provides data communication services through the worldwide packet data communication network, now commonly referred to as the "Internet” 728.
  • Internet 728 uses electrical, electromagnetic or optical signals that carry digital data streams.
  • the signals through the various networks and the signals on network link 120 and through communication interface 718, which carry the digital data to and from computer system 100, are exemplary forms of carrier waves transporting the information.
  • Computer system 100 can send messages and receive data, including program code, through the network(s), network link 120, and communication interface 718.
  • a server 130 might transmit a requested code for an application program through Internet 728, ISP 726, local network 722 and communication interface 718.
  • one such downloaded application provides for constructing TSVs and performing various semantic operations as described herein.
  • the received code may be executed by processor 704 as it is received, and/or stored in storage device 710, or other non-volatile storage for later execution. In this manner, computer system 100 may obtain application code in the form of a carrier wave.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Computational Linguistics (AREA)
  • Data Mining & Analysis (AREA)
  • Databases & Information Systems (AREA)
  • Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)

Abstract

L'invention porte sur un système et un procédé pour identifier un ou plusieurs ensembles de données, tels que des publicités, qui sont associés contextuellement à un ensemble de données objet, tel qu'une page Internet qui est examinée par un utilisateur, sur la base d'analyses de vecteurs sémantiques uniques, tels que des vecteurs sémantiques pouvant être entraînés (TSV), qui représentent la page Internet et les publicités, et de représentations sémantiques comprenant des informations de mots-clés représentatifs des publicités et de la page Internet.
PCT/US2008/071505 2008-07-29 2008-07-29 Procédé et appareil pour associer des ensembles de données à l’aide de vecteurs sémantiques et d'analyses de mots-clés WO2010014082A1 (fr)

Priority Applications (4)

Application Number Priority Date Filing Date Title
JP2011521074A JP2011529600A (ja) 2008-07-29 2008-07-29 意味ベクトルおよびキーワード解析を使用することによるデータセットを関係付けるための方法および装置
EP08782506A EP2307951A4 (fr) 2008-07-29 2008-07-29 Procédé et appareil pour associer des ensembles de données à l aide de vecteurs sémantiques et d'analyses de mots-clés
PCT/US2008/071505 WO2010014082A1 (fr) 2008-07-29 2008-07-29 Procédé et appareil pour associer des ensembles de données à l’aide de vecteurs sémantiques et d'analyses de mots-clés
CN200880001312A CN101802776A (zh) 2008-07-29 2008-07-29 应用语义向量和关键字分析关联数据集的方法和装置

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
PCT/US2008/071505 WO2010014082A1 (fr) 2008-07-29 2008-07-29 Procédé et appareil pour associer des ensembles de données à l’aide de vecteurs sémantiques et d'analyses de mots-clés

Publications (1)

Publication Number Publication Date
WO2010014082A1 true WO2010014082A1 (fr) 2010-02-04

Family

ID=41610613

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/US2008/071505 WO2010014082A1 (fr) 2008-07-29 2008-07-29 Procédé et appareil pour associer des ensembles de données à l’aide de vecteurs sémantiques et d'analyses de mots-clés

Country Status (4)

Country Link
EP (1) EP2307951A4 (fr)
JP (1) JP2011529600A (fr)
CN (1) CN101802776A (fr)
WO (1) WO2010014082A1 (fr)

Cited By (12)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2012093863A (ja) * 2010-10-26 2012-05-17 Yahoo Japan Corp 広告選択装置、方法及びプログラム
US20140032539A1 (en) * 2012-01-10 2014-01-30 Ut-Battelle Llc Method and system to discover and recommend interesting documents
JP2014137620A (ja) * 2013-01-15 2014-07-28 Yahoo Japan Corp 情報配信装置、及び、情報配信方法
CN105022754A (zh) * 2014-04-29 2015-11-04 腾讯科技（深圳）有限公司 基于社交网络的对象分类方法及装置
US9195470B2 (en) 2013-07-22 2015-11-24 Globalfoundries Inc. Dynamic data dimensioning by partial reconfiguration of single or multiple field-programmable gate arrays using bootstraps
US10360520B2 (en) 2015-01-06 2019-07-23 International Business Machines Corporation Operational data rationalization
US10643031B2 (en) 2016-03-11 2020-05-05 Ut-Battelle, Llc System and method of content based recommendation using hypernym expansion
CN113330474A (zh) * 2019-06-26 2021-08-31 谷歌有限责任公司 用于提供内容候选的系统和方法
CN113449111A (zh) * 2021-08-31 2021-09-28 苏州工业园区测绘地理信息有限公司 基于时空语义知识迁移的社会治理热点话题自动识别方法
CN113609264A (zh) * 2021-06-28 2021-11-05 国网北京市电力公司 电力系统节点的数据查询方法、装置
CN114187605A (zh) * 2021-12-13 2022-03-15 苏州方兴信息技术有限公司 一种数据集成方法、装置和可读存储介质
WO2024074760A1 (fr) * 2022-10-04 2024-04-11 Thirdpresence Oy Agencement de gestion de contenu

Families Citing this family (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103649905B (zh) * 2011-03-10 2015-08-05 特克斯特怀茨有限责任公司 用于统一信息表示的方法和系统及其应用
JP6228425B2 (ja) * 2013-10-25 2017-11-08 株式会社Ｎｔｔドコモ 広告生成装置および広告生成方法
JP6583686B2 (ja) * 2015-06-17 2019-10-02 パナソニックＩｐマネジメント株式会社 意味情報生成方法、意味情報生成装置、およびプログラム
EP3506160B1 (fr) * 2017-12-28 2022-06-01 Dassault Systèmes Segmentation sémantique de plans de sol 2d à l'aide d'un classificateur au niveau du pixel
CN109558586B (zh) * 2018-11-02 2023-04-18 中国科学院自动化研究所 一种资讯的言据自证评分方法、设备和存储介质
CN111199259B (zh) * 2018-11-19 2023-06-20 中国电信股份有限公司 标识转换方法、装置和计算机可读存储介质

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6134532A (en) * 1997-11-14 2000-10-17 Aptex Software, Inc. System and method for optimal adaptive matching of users to most relevant entity and information in real-time
US20050216516A1 (en) * 2000-05-02 2005-09-29 Textwise Llc Advertisement placement method and system using semantic analysis
US7089194B1 (en) * 1999-06-17 2006-08-08 International Business Machines Corporation Method and apparatus for providing reduced cost online service and adaptive targeting of advertisements

Family Cites Families (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP3195752B2 (ja) * 1997-02-28 2001-08-06 シャープ株式会社 検索装置
JP2005173795A (ja) * 2003-12-09 2005-06-30 Canon Inc 文書検索装置、およびその検索方法、並びに記憶媒体
JP2005326970A (ja) * 2004-05-12 2005-11-24 Mitsubishi Electric Corp 構造化文書曖昧検索装置及びそのプログラム
JP4728125B2 (ja) * 2006-01-11 2011-07-20 ヤフー株式会社 索引ファイルを用いた文書検索の方法、索引ファイルを用いた文書検索サーバ、及び索引ファイルを用いた文書検索プログラム
CN100517330C (zh) * 2007-06-06 2009-07-22 华东师范大学 一种基于语义的本地文档检索方法

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6134532A (en) * 1997-11-14 2000-10-17 Aptex Software, Inc. System and method for optimal adaptive matching of users to most relevant entity and information in real-time
US7089194B1 (en) * 1999-06-17 2006-08-08 International Business Machines Corporation Method and apparatus for providing reduced cost online service and adaptive targeting of advertisements
US20050216516A1 (en) * 2000-05-02 2005-09-29 Textwise Llc Advertisement placement method and system using semantic analysis

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
See also references of EP2307951A4 *

Cited By (17)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2012093863A (ja) * 2010-10-26 2012-05-17 Yahoo Japan Corp 広告選択装置、方法及びプログラム
US20140032539A1 (en) * 2012-01-10 2014-01-30 Ut-Battelle Llc Method and system to discover and recommend interesting documents
US9558185B2 (en) * 2012-01-10 2017-01-31 Ut-Battelle Llc Method and system to discover and recommend interesting documents
JP2014137620A (ja) * 2013-01-15 2014-07-28 Yahoo Japan Corp 情報配信装置、及び、情報配信方法
US9195470B2 (en) 2013-07-22 2015-11-24 Globalfoundries Inc. Dynamic data dimensioning by partial reconfiguration of single or multiple field-programmable gate arrays using bootstraps
US9875294B2 (en) 2014-04-29 2018-01-23 Tencent Technology (Shenzhen) Company Limited Method and apparatus for classifying object based on social networking service, and storage medium
WO2015165372A1 (fr) 2014-04-29 2015-11-05 Tencent Technology (Shenzhen) Company Limited Procédé et appareil pour classifier un objet sur la base d'un service de réseautage social, et support de stockage
EP3138058A4 (fr) * 2014-04-29 2017-03-08 Tencent Technology (Shenzhen) Company Limited Procédé et appareil pour classifier un objet sur la base d'un service de réseautage social, et support de stockage
CN105022754A (zh) * 2014-04-29 2015-11-04 腾讯科技（深圳）有限公司 基于社交网络的对象分类方法及装置
US10360520B2 (en) 2015-01-06 2019-07-23 International Business Machines Corporation Operational data rationalization
US10572838B2 (en) 2015-01-06 2020-02-25 International Business Machines Corporation Operational data rationalization
US10643031B2 (en) 2016-03-11 2020-05-05 Ut-Battelle, Llc System and method of content based recommendation using hypernym expansion
CN113330474A (zh) * 2019-06-26 2021-08-31 谷歌有限责任公司 用于提供内容候选的系统和方法
CN113609264A (zh) * 2021-06-28 2021-11-05 国网北京市电力公司 电力系统节点的数据查询方法、装置
CN113449111A (zh) * 2021-08-31 2021-09-28 苏州工业园区测绘地理信息有限公司 基于时空语义知识迁移的社会治理热点话题自动识别方法
CN114187605A (zh) * 2021-12-13 2022-03-15 苏州方兴信息技术有限公司 一种数据集成方法、装置和可读存储介质
WO2024074760A1 (fr) * 2022-10-04 2024-04-11 Thirdpresence Oy Agencement de gestion de contenu

Also Published As

Publication number Publication date
EP2307951A4 (fr) 2012-12-19
JP2011529600A (ja) 2011-12-08
EP2307951A1 (fr) 2011-04-13
CN101802776A (zh) 2010-08-11

Similar Documents

Publication Publication Date Title
US7912868B2 (en) Advertisement placement method and system using semantic analysis
EP2307951A1 (fr) Procédé et appareil pour associer des ensembles de données à l aide de vecteurs sémantiques et d'analyses de mots-clés
US9317498B2 (en) Systems and methods for generating summaries of documents
US8051080B2 (en) Contextual ranking of keywords using click data
US8606808B2 (en) Finding relevant documents
US9058394B2 (en) Matching and recommending relevant videos and media to individual search engine results
US8768960B2 (en) Enhancing keyword advertising using online encyclopedia semantics
CN104885081B (zh) 搜索系统和相应方法
US20090254540A1 (en) Method and apparatus for automated tag generation for digital content
US20130060769A1 (en) System and method for identifying social media interactions
Xu et al. Web content mining
US20080168056A1 (en) On-line iterative multistage search engine with text categorization and supervised learning
WO2006108069A2 (fr) Recherche de contenu accessible par l'intermediaire de formulaires web
WO2008097856A2 (fr) Moteur de distribution de résultats de recherche
WO2014107801A1 (fr) Procédés et appareils permettant d'identifier des concepts correspondant à des informations d'entrée
US20100094826A1 (en) System for resolving entities in text into real world objects using context
CN112989208B (zh) 一种信息推荐方法、装置、电子设备及存储介质
KR20080037413A (ko) 온라인 문맥기반 광고 장치 및 방법
US20220358122A1 (en) Method and system for interactive keyword optimization for opaque search engines
Shah et al. DOM-based keyword extraction from web pages
WO2012091541A1 (fr) Système de constructeur de web sémantique et procédé pour ce système
Murfi et al. A two-level learning hierarchy of concept based keyword extraction for tag recommendations
Tsapatsoulis Web image indexing using WICE and a learning-free language model
Zheng et al. An improved focused crawler based on text keyword extraction
Tarannum et al. NLP based Text Summarization Techniques for News Articles: Approaches and Challenges

Legal Events

Date Code Title Description
WWE Wipo information: entry into national phase

Ref document number: 200880001312.7

Country of ref document: CN

WWE Wipo information: entry into national phase

Ref document number: 2593/CHENP/2009

Country of ref document: IN

WWE Wipo information: entry into national phase

Ref document number: 2008782506

Country of ref document: EP

ENP Entry into the national phase

Ref document number: 2011521074

Country of ref document: JP

Kind code of ref document: A

121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 08782506

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE