WO2015084757A1 - Systèmes et procédés de traitement de données stockées dans une base de données - Google Patents

Systèmes et procédés de traitement de données stockées dans une base de données Download PDF

Info

Publication number
WO2015084757A1
WO2015084757A1 PCT/US2014/067994 US2014067994W WO2015084757A1 WO 2015084757 A1 WO2015084757 A1 WO 2015084757A1 US 2014067994 W US2014067994 W US 2014067994W WO 2015084757 A1 WO2015084757 A1 WO 2015084757A1
Authority
WO
WIPO (PCT)
Prior art keywords
computer
fact
search query
features
identifier
Prior art date
Application number
PCT/US2014/067994
Other languages
English (en)
Inventor
Rakesh Dave
Sanjay Boddhu
Scott Lightner
Robert Flagg
Original Assignee
Qbase, 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 Qbase, LLC filed Critical Qbase, LLC
Publication of WO2015084757A1 publication Critical patent/WO2015084757A1/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
    • G06F16/35—Clustering; Classification
    • 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
    • G06F16/3332—Query translation
    • G06F16/3334—Selection or weighting of terms from queries, including natural language queries
    • 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
    • G06F16/35—Clustering; Classification
    • G06F16/355—Class or cluster creation or modification

Definitions

  • the present disclosure relates in general to information data mining from document sources, and more specifically to extraction of facts from documents.
  • the present disclosure relates in general to information data, and more specifically to a method for building a knowledge base storage of feature co-occurrences.
  • the present disclosure relates in general to in-memory databases, and more specifically to search methods for discovering and exploring feature knowledge within in-memory databases.
  • Electronic document corpora may contain vast amounts of information. For a person searching for specific information in a document corpus, identifying key information may be troublesome. Manually crawling each document and highlighting or extracting important information may even be impossible depending on the size of the document corpus. At times a reader may only be interested in facts or asserted information.
  • intelligent computer systems for extracting features in an automated matter may be commonly used to aid in fact extraction. However, current intelligent systems fail to properly extract facts and associate them with other extracted features such as entities, topics, events and other feature types.
  • Searching information about entities i.e., people, locations, organizations
  • entities i.e., people, locations, organizations
  • searching information about entities may often be ambiguous, which may lead to imprecise text processing functions and thus imprecise data analysis.
  • a reference to "Paris” could refer to a city in the country of France, cities in the States of Texas, Tennessee or Illinois, or even a person (e.g., "Paris Hilton").
  • Associating entities with co-occurring features may prove helpful in disambiguating different entities.
  • the system includes an entity extraction computer module used to extract and disambiguate independent entities from an electronic document, such as a text file.
  • the system may further include a topic extractor computer module configured to determine a topic related to the text file.
  • the system may extract possible facts described in the text by comparing text string structures against a fact template store.
  • the fact template store may be built by revising documents containing facts and recording a commonly used fact sentence structure.
  • the extracted facts may then be associated with extracted entities and topics to determine a confidence score that may serve as an indication of the accuracy of the fact extraction.
  • a method comprises receiving, by an entity extraction computer, an electronic document having unstructured text and extracting, by the entity extraction computer, an entity identifier from the unstructured text in the electronic document.
  • the method further includes extracting, by a topic extraction computer, a topic identifier from the unstructured text in the electronic document, and extracting, by a fact extraction computer, a fact identifier from the unstructured text in the electronic document by comparing text string structures in the unstructured text to a fact template database, the fact template database having stored therein a fact template model identifying keywords pertaining to specific fact identifiers and corresponding keyword weights.
  • the method further includes associating, by a fact relatedness estimator computer, the entity identifier with the topic identifier and the fact identifier to determine a confidence score indicative of a degree of accuracy of extraction of the fact identifier.
  • a system comprising one or more server computers having one or more processors executing computer readable instructions for a plurality of computer modules.
  • the computer modules include an entity extraction module configured to receive an electronic document having unstructured text and extract an entity identifier from the unstructured text in the electronic document, a topic extraction module configured to extract a topic identifier from the unstructured text in the electronic document, and a fact extraction module configured to extract a fact identifier from the unstructured text in the electronic document by comparing text string structures in the unstructured text to a fact template database, the fact template database having stored therein a fact template model identifying keywords pertaining to specific fact identifiers and corresponding keyword weights.
  • the system further includes a fact relatedness estimator module configured to associate the entity identifier with the topic identifier and the fact identifier to determine a confidence score indicative of a degree of accuracy of extraction of the fact identifier.
  • a non-transitory computer readable medium having stored thereon computer executable instructions.
  • the instructions comprise receiving, by an entity extraction computer, an electronic document having unstructured text, extracting, by the entity extraction computer, an entity identifier from the unstructured text in the electronic document, and extracting, by a topic extraction computer, a topic identifier from the unstructured text in the electronic document.
  • the instructions further include extracting, by a fact extraction computer, a fact identifier from the unstructured text in the electronic document by comparing text string structures in the unstructured text to a fact template database, the fact template database having stored therein a fact template model identifying keywords pertaining to specific fact identifiers and corresponding keyword weights, and associating, by a fact relatedness estimator computer, the entity identifier with the topic identifier and the fact identifier to determine a confidence score indicative of a degree of accuracy of extraction of the fact identifier.
  • the system may score records against the one or more queries, where the system may score the match of one or more available fields of the records and may then determine a score for the overall match of the records.
  • the system may determine whether the score is above a predefined acceptance threshold, where the threshold may be defined in the search query or may be a default value.
  • fuzzy matching algorithms may compare records temporarily stored in collections with the one or more queries being generated by the system.
  • numerous analytics computer modules may be plugged to the in-memory data base and the user may be able to modify the relevant analytical parameters of each analytics computer module through a user interface.
  • a method comprises receiving, by a search manager computer, a search query from a user computing device configured to receive a selection, from a user, of an analytic computer that processes search query results for presentation to the user.
  • the method further includes submitting, by the search manager computer, the search query to a search conductor computer for conducting a search, receiving, by the search manager computer, the search query results from the search conductor computer, the search query results having one or more records matching the search query, and forwarding, by the search manager computer, the search query results to the analytic computer selected by the user to process the search query results for the presentation.
  • the method also includes receiving, by the search manager computer, the search query results processed by the analytic computer selected by the user, and returning, by the search manager computer, the search query results to the user device for the presentation to the user in accordance with the processing.
  • a system comprising one or more server computers having one or more processors executing computer readable instructions for a plurality of computer modules including a search manager computer module configured to receive a search query from a user computing device that is configured to receive a selection, from a user, of an analytic computer module that processes search query results for presentation to the user.
  • a search manager computer module configured to receive a search query from a user computing device that is configured to receive a selection, from a user, of an analytic computer module that processes search query results for presentation to the user.
  • the search manager computer module is further configured to: submit the search query to a search conductor computer module configured to conduct a search, receive the search query results from the search conductor computer module, the search query results having one or more records matching the search query, forward the search query results to the analytic computer module selected by the user to process the search query results for the presentation, receive the search query results processed by the analytic computer module selected by the user, and return the search query results to the user device for the presentation to the user in accordance with the processing.
  • a system and method for building a knowledge base of feature co-occurrences from a document corpus may include a plurality of software and hardware computer modules to extract different features such as entities, topics, events, facts and/or any other features that may be derived from a document.
  • the system may crawl each document in a document corpus and extract features from each individual document. After different features are extracted from a document they may be submitted to a knowledge base aggregator where the co-occurrences of two or more features may be aggregated with co- occurrences of the same features in different documents. Once the aggregation for the co- occurring features reach a determined threshold the co-occurrences and additional metadata related to the co-occurring features may be stored in the knowledge base.
  • the knowledge base of co-occurring features may serve to assist in subsequent disambiguation of features.
  • the knowledge base may be created using a single document corpus or a plurality of document corpora.
  • a method includes crawling, via an entity extraction computer, the corpus of electronic documents, extracting, via the entity extraction computer, a plurality of features from each of the crawled documents in the corpus and aggregating, via a knowledge base aggregator computer, instances of co-occurrence of two or more of the plurality of features across the crawled documents to determine a count of the instances of co-occurrence.
  • the method further includes adding, via the knowledge base aggregator computer, an instance of co-occurrence of the two or more features to the feature co-occurrence database when the count of the instances of co-occurrence exceeds a predetermined threshold.
  • a system in another embodiment, includes an entity extraction computer configured to crawl a corpus of electronic documents and extract a plurality of features from each of the crawled documents in the corpus.
  • the system further includes a knowledge base aggregator computer configured to aggregate instances of cooccurrence of two or more of the plurality of features across the crawled documents to determine a count of the instances of co-occurrence, and add an instance of co-occurrence of the two or more features to a feature co-occurrence database when the count of the instances of co-occurrence exceeds a predetermined threshold.
  • a non-transitory computer readable medium having stored thereon computer executable instructions comprises crawling, via an entity extraction computer, a corpus of electronic documents; extracting, via the entity extraction computer, a plurality of features from each of the crawled documents in the corpus; aggregating, via a base aggregator computer, instances of co-occurrence of two or more of the plurality of features across the crawled documents to determine a count of the instances of co-occurrence; and adding, via the base aggregator computer, an instance of co-occurrence of the two or more features to a feature co-occurrence database when the count of the instances of co-occurrence exceeds a predetermined threshold.
  • FIG. 1 is a diagram of a fact extraction system, according to an embodiment.
  • FIG. 2 is diagram of a system for training a fact concept store, according to an embodiment.
  • FIG. 3 is a flow chart of a method for building a fact template store of FIG. 2, according to an embodiment.
  • FIG. 4 is a flowchart of a search method for discovering and exploring feature knowledge, according to an embodiment.
  • FIG. 5 is a flowchart of process executed by a link on-the-fly module, according to an embodiment.
  • FIG. 6 is a system employed for disambiguating features according to an exemplary embodiment.
  • FIG. 7 is a diagram of a central computer server system for building a knowledge base of co-occurrences, according to an embodiment.
  • FIG. 8 is a diagram of a co-occurring aggregation method, according to an embodiment.
  • Entity Extraction refers to information processing methods for extracting information such as names, places, and organizations.
  • Corpus refers to a collection of one or more documents
  • Event Concept Store refers to a database of Event template models.
  • Event refers to one or more features characterized by at least the features' occurrence in real-time.
  • Event Model refers to a collection of data that may be used to compare against and identify a specific type of event.
  • Module refers to a computer or software components suitable for carrying out at least one or more tasks.
  • Fracts refers to asserted information about features found in an electronic document.
  • Document refers to a discrete electronic representation of information having a start and end.
  • Facet refers to clearly defined, mutually exclusive, and collectively exhaustive aspects, properties or characteristics of a class, specific subject, topic or feature.
  • Knowledge base refers to a computer database containing disambiguated features or facets.
  • Live corpus refers to a corpus that is constantly fed as new electronic documents are uploaded into a network.
  • Memory refers to any hardware component suitable for storing information and retrieving said information at a sufficiently high speed.
  • Analytics Parameters refers to parameters that describe the operation that an analytic computer module may have to perform in order to get specific results.
  • Link on-the-fly module refers to a linking computer module that updates data as a live corpus is updated.
  • Node refers to a computer hardware configuration suitable for running one or more modules.
  • Node Cluster refers to a set of one or more nodes.
  • Query refers to an electronic request to retrieve information from one or more suitable databases.
  • Record refers to one or more pieces of information that may be handled as a unit.
  • Collection refers to a discrete set of records.
  • Partition refers to an arbitrarily delimited portion of records of a collection.
  • Prefix refers to a string of a given length that includes the longest string of key characters shared by all subtrees of the node and a data record field for storing a reference to a data record.
  • Database refers to any computer system including any combination of node clusters and computer modules suitable for storing one or more collections and suitable to process one or more queries.
  • Analytics Agent or “Analytics Module” refers to a computer or computer module configured to at least receive one or more records, process said one or more records, and return the resulting one or more processed records.
  • Search Manager or " SM” refers to a computer or computer module configured to at least receive one or more queries and return one or more search results.
  • Search Conductor or “SC” refers to a computer or computer module configured to at least run one or more search queries on a partition and return the search results to one or more search managers.
  • Standard refers to subjective assessments associated with a document, part of a document, or feature.
  • Topicic refers to a set of thematic information which is at least partially derived from a corpus.
  • Feature attribute refers to metadata associated with a feature; for example, location of a feature in a document, confidence score, among others.
  • Sources may include news sources, social media websites and/or any sources that may include data pertaining to events.
  • FIG. 1 depicts an embodiment of a system 100 for extracting facts from an electronic document.
  • Embodiments of the disclosed system may be implemented in various operating environments that include personal computers, server computers, handheld or laptop devices, multiprocessor systems, microprocessor-based systems, programmable consumer electronics, network PCs, minicomputers, mainframe computers, and distributed computing environments.
  • the document corpus computer module 102 may provide an input of an electronic document containing unstructured text such as, for example, a news feed article, a file from a digital library, a blog, a forum, a digital book and/or any file containing natural language text.
  • unstructured text such as, for example, a news feed article, a file from a digital library, a blog, a forum, a digital book and/or any file containing natural language text.
  • the process may involve crawling through document file received from the corpus 102.
  • An electronic document may include information in unstructured text format which may be crawled using natural language processing techniques (LP).
  • LP natural language processing techniques
  • Some NLP techniques include, for example, removing stop words, tokenization, stemming and part-of speech tagging among others known in the art.
  • An individual file may first go through an entity extraction computer module
  • Entity extraction module 104 where entities (e.g., a person, location, or organization name) are identified and extracted.
  • Entity extraction module 104 may also include disambiguation methods which may differentiate ambiguous entities. Disambiguation of entities may be performed in order to attribute a fact to an appropriate entity.
  • a method for entity disambiguation may include, for example, comparing extracted entities and co-occurrences with other entities or features against a knowledge base of co-occurring features in order to identify specific entities the document may be referring to. Other methods for entity disambiguation may also be used and are included within the scope of this disclosure.
  • entity extraction computer module 104 may be implemented as a hardware and/or software module in a single computer or in a distributed computer architecture.
  • Topic extractor module 108 may extract the theme or topic of a single document file. In most cases a file may include a single topic, however a plurality of topics may also exist in a single document. Topic extraction techniques may include, for example, comparing keywords against models built with a multi-component extension of latent Dirichlet allocation (MC- LDA), among other techniques for topic identification. A topic may then be appended to a fact in order to provide more accurate information.
  • MC- LDA multi-component extension of latent Dirichlet allocation
  • System 100 may include a fact extractor computer module 112.
  • Fact extractor module 112 may be a hardware and/or software computer module executing programmatic logic that may extract facts by crawling through the document.
  • Fact extractor module 112 may compare text structures against fact models stored in a fact template store 114 in order to extract and determine the probability of an extracted fact and the associated fact type.
  • a fact relatedness estimator computer module 116 may correlate all features in order to determine a fact relation to other features and assign a confidence score that may serve as an indication that an extracted fact is accurate.
  • Fact relatedness estimator module 116 may calculate a confidence score based on a text distance between parts of text from where a fact was extracted and where a topic or entity was extracted. For example, consider the fact example "President said the bill will pass” extracted from a document where the identified topic was "immigration”. Fact relatedness estimator module 116 may measure the distances between the fact sentence "President said the bill will pass” and the sentence from where the topic "immigration" was extracted.
  • the fact relatedness estimator module 116 may also calculate confidence score by comparing co-occurring entities in the same document file. For example, considering the same example used before the entity "president" may be mentioned at different parts in the document. A co-occurrence of an entity mentioned in a fact with the same entity in a different part of the document may increase a confidence score associated with the fact.
  • the distances between co-occurring entities in relation to facts may also be used in determining confidences scores. Distances in text may be calculated using methods such as tokenization or any other NLP methods.
  • Verified fact store 118 may be a computer database used by various applications in order to query for different facts associated with the purpose of a given application.
  • FIG. 1 illustrates an exemplary embodiment and is in no way limiting the scope of the invention. Additional modules for extracting different features not illustrated in FIG. 1 may also be included and are to be considered within the scope of the invention. As those of skill in the art will realize, all hardware and software modules described in the present disclosure may be implemented in a single special purpose computer or in a distributed computer architecture across a plurality of special purpose computers.
  • FIG. 2 is an embodiment of a training computer system 200 for building a fact template store 214.
  • a plurality of documents 202 may be tagged, for example by a computer process, in order to identify key words pertaining to specific facts and assign weights to those keywords.
  • an embodiment of a fact template model 206 may be "The President said the bill will pass.”
  • the tagging process of the system 200 can identify, tag and record the sentence structure of the fact.
  • the person may identify the keyword "said" preceded by an entity (e.g., the "President") and proceeded by some string (e.g., "the bill will pass") which may represent the value of the fact.
  • the model may then be stored in fact template store 214 along with metadata such as for example, a count of how many times that sentence structure is repeated across different documents, a fact type classification, a confidence score that serves as an indication of how strongly the sentence structure may resemble a fact.
  • Fact template models 206 may be used in subsequent text comparisons in order to extract facts from document files.
  • FIG. 3 is an embodiment of a method for building a fact template store of
  • FIG. 2 the computer system 200 (FIG. 2) tags electronic documents in a corpus of documents to identify keywords pertaining to facts.
  • the system 200 assigns weights to tagged keywords.
  • the system 200 selects a fact template model having the identified keywords (from other electronic documents in the corpus) and stores the fact template in the fact template store database along with the metadata, as discussed above in connection with FIG. 2.
  • the fact template model is used in text comparisons in the process of fact extraction, as discussed in FIG. 1 above.
  • FIG. 4 is a flow chart describing a search method 400 for discovering and exploring feature knowledge, according to an embodiment.
  • the process may start when a user generates a search query, step 402.
  • One or more user workstations e.g., personal computer, smartphone, tablet computer, mobile device, or the like
  • the user interfaces may receive from a user workstation a selection of an option of one or more of a set of analytic methods that may be applied to the results of the search query.
  • the user workstations may also allow for the selection of thresholds of acceptance of different levels (e.g., of search query results).
  • these queries and thresholds can be generated automatically, may be transmitted from a computing device, or may be predetermined.
  • the query may be received, in step 404, by one or more search manager computer modules (SM) embodied on a computer readable medium and executed by a processor.
  • the one or more queries generated by the interaction of one or more users with one or more user interfaces may be received by one or more search manager computer modules.
  • the queries may be represented in a markup language, including XML and HTML.
  • the queries may be represented in a data structure, including embodiments where the queries are represented in JSON.
  • a query may be represented in compact or binary format.
  • the received queries may be parsed by the one or more SM computer modules, in step 406. This process may allow the system to determine if field processing is desired, in step 408. In one or more embodiments, the system may be capable of determining if the process is required using information included in the query. In one or more other embodiments, the one or more search manager computer modules may automatically or dynamically determine which one or more fields may undergo a desired processing.
  • the one or more SM computer modules may apply one or more suitable processing techniques to the one or more desired fields, during the search manager processes fields step 410.
  • suitable processing techniques may include address standardization, proximity boundaries, and nickname interpretation, among others.
  • suitable processing techniques may include the extraction of prefixes from strings and the generation of non-literal keys that may later be used to apply fuzzy matching techniques.
  • the one or more SM computer modules construct the search query, in step 412, they may construct additional search queries associated with the current search query.
  • the search queries may be constructed so as to be processed as a stack-based search.
  • one or more SM computer modules may send search query to one or more search conductor computer modules (SC), in step 414, where said one or more SC computer modules may be associated with collections specified in the one or more search queries.
  • SC search conductor computer modules
  • the one or more search conductors may score records against the one or more queries, where the search conductors may score the match of one or more fields of the records and may then determine a score for the overall match of the records with the one or more queries.
  • the system may determine whether the score is above a predefined acceptance threshold, where the threshold may be defined in the search query or may be a default value. In one or more embodiments, the default score thresholds may vary according to the one or more record fields being scored. If the SC computer module determines that the scores are above the desired threshold, the records may be added to a results list. The SC computer module may continue to score records until it determines that a record is the last in the partition. If the SC computer module determines that the last record in a partition has been processed, the SC computer module may then sort the resulting results list. The DC computer module may then return the results list to a SM computer module.
  • SM computer module When SM computer module receives and collates search results from SC computer modules, step 416, the one or more search conductors return the one or more search results to the one or more search managers; where, in one or more embodiments, said one or more search results may be returned asynchronously.
  • the one or more SM may then compile results from the one or more SC computer modules into one or more results lists.
  • the one or more SM computer modules may automatically determine which one or more fields may undergo one or more desired analytic processes. Then, the one or more SM computer modules may send the search results to analytic computer modules, in step 418.
  • the one or more results lists compiled by one or more SM computer modules may be sent to one or more analytics agent computers, where each analytics agent computer may include one or more analytics computer modules configured to execute a corresponding one of the one or more suitable processing techniques.
  • suitable techniques may include rolling up several records into a more complete record, performing one or more analytics on the results, and determining information about neighboring records, amongst others.
  • analytics agent computers may execute disambiguation computer modules, link computer modules, link on-the-fly computer modules, or other suitable computer modules and corresponding algorithms.
  • the system may allow for user workstation to customize the analytics modules according to particular inputs.
  • the one or more analytics agents may return one or more processed results lists, step 420, to the one or more SM computer modules.
  • a SM computer module may return search results to the user device's user interface, step 422.
  • the one or more SM computer modules may decompress the one or more results list and return them to the user interface that initiated the query.
  • the search results may be temporarily stored in a knowledge base database and returned to the user interface of the user computing device (e.g., workstation).
  • the knowledge base may be used to temporarily store clusters of relevant disambiguated features.
  • MEMDB in-memory database
  • the new disambiguated set of features may be compared with the existing knowledge base in order to determine the relationship between features and automatically determine if there is a match between the new features and previously extracted features.
  • the knowledge base may be automatically updated and the identification (ID) of the matching features may be returned. If the features compared do not match with any of the already extracted features, a unique ID is assigned to the disambiguated entity or feature, and the ID is associated with the cluster of defining features and stored within the knowledge base of the MEMDB.
  • the user computing device may determine if a query needs further modification, in step 424, to achieve the desired results. If the desired results are achieved, the process may end, in step 426. If the desired results are not achieved, the user computing device may generate a new query by changing the type of analytics desired (e.g., by selecting a different analytics computer module executing a different analytics algorithm) or the level of precision and the user computing device may adjust how knowledge is linked to find stronger or looser relationships. In some embodiments, a new search may be generated and combined it with a current search.
  • FIG. 5 is a flowchart of a process 500 executed by a link OTF computer sub- module, which may be employed for disambiguating features in the search method 400 (FIG.4), according to an embodiment.
  • Link OTF sub-module may be capable of constantly evaluating, scoring, linking, and clustering a feed of information.
  • Link OTF sub-module may perform dynamic records linkage using multiple algorithms.
  • search results may be constantly fed into the link OTF computer sub-module.
  • the input of data may be followed by a match scoring algorithm application, step 504, where one or more match scoring algorithms may be applied simultaneously in multiple search nodes of the MEMDB while performing fuzzy key searches for evaluating and scoring the relevant results, taking in account multiple feature attributes, such as string edit distances, phonetics, and sentiments, among others.
  • a match scoring algorithm application step 504 where one or more match scoring algorithms may be applied simultaneously in multiple search nodes of the MEMDB while performing fuzzy key searches for evaluating and scoring the relevant results, taking in account multiple feature attributes, such as string edit distances, phonetics, and sentiments, among others.
  • Linking algorithm application step 506 may be added to compare all candidate records, identified during match scoring algorithm application step 504, to each other.
  • Linking algorithm application step 506 may include the use of one or more analytical linking algorithms capable of filtering and evaluating the scored results of the fuzzy key searches performed inside the multiple search nodes of the MEMDB.
  • cooccurrence of two or more features across the collection of identified candidate records in the MEMDB may be analyzed to improve the accuracy of the process.
  • Different weighted models and confidence scores associated with different feature attributes may be taken into account for linking algorithm application 506.
  • the linked results may be arranged in clusters of related features and returned to the user interface, as part of return of linked records clusters, step 508.
  • FIG. 6 is an illustrative diagram of an embodiment of a system 600 for disambiguating features in unstructured text and including the link OTF sub-module 612 discussed above in connection with FIG.5.
  • the system 600 hosts an in-memory database and comprises one or more nodes.
  • the system 600 includes one or more processors executing computer instructions for a plurality of special-purpose computer modules 601, 602, 608, 611, 612, and 614 to disambiguate features within one or more documents.
  • the document input modules 601, 602 receive documents from internet based sources and/or a live corpus of documents.
  • a large number of new documents may be uploaded substantially simultaneously from a user workstation 606 or other computing device into the document input module 602 through a network connection (NC) 604. Therefore, the source may be constantly receiving an input of new knowledge, using updated information provided by user workstations 606, where such new knowledge is not pre-linked in a static way.
  • the number of documents to be evaluated may be infinitely increasing.
  • the system 600 is therefore able to process large volumes of documents in a more efficient manner to discover and explore feature knowledge.
  • An in-memory database (MEMDB) computer 608 may facilitate a faster disambiguation process, such as by executing a disambiguation process on-the-fly, which may facilitate reception of the latest information that is going to contribute to MEMDB 608.
  • Various methods for linking the features may be employed, which may essentially use a weighted model for determining which entity types are most important, which have more weight, and, based on confidence scores, determine how confident the extraction and disambiguation of the correct features has been performed, and that the correct feature may go into the resulting cluster of features. As shown in FIG. 6, as more system nodes are working in parallel, the process may become more efficient.
  • an extraction module 611 when a new document arrives into the system 600 via the document input module 601, 602 through a network connection 604, an extraction module 611 performs feature extraction and, then, a feature disambiguation sub- module 614 may perform feature disambiguation on the new document.
  • Extraction module 611 and feature disambiguation module 614 are components of system 600.
  • extraction module 611 and disambiguation module 614 are separate modules of the system 600, though extraction module 611 and disambiguation module 614 can be configured as a single module, hosted on a single computer, or each can be configured as a separate computer. In one configuration, extraction module 614 and disambiguation module 614 may each be executed by the MEMDB 608.
  • the extracted new features 610 may be included in the MEMDB 608 to pass through link OTF sub-module 612; where the features may be compared and linked, and a feature ID of disambiguated feature 610 may be returned to the user workstation 606 as a result from a query.
  • the resulting feature cluster defining the disambiguated feature may optionally be returned to the user workstation 606.
  • MEMDB computer 608 can be a database storing data in records controlled by a database management system (DBMS) (not shown) configured to store data records in a device's main memory, as opposed to conventional databases and DBMS modules that store data in "disk” memory.
  • DBMS database management system
  • Conventional disk storage requires processors (CPUs) to execute read and write commands to a device's hard disk, thus requiring CPUs to execute instructions to locate (i.e., seek) and retrieve the memory location for the data, before performing some type of operation with the data at that memory location.
  • In-memory database systems access data that is placed into main memory, and then addressed accordingly, thereby mitigating the number of instructions performed by the CPUs and eliminating the seek time associated with CPUs seeking data on hard disk.
  • In-memory databases may be implemented in a distributed computing architecture, which may be a computing system comprising one or more nodes configured to aggregate the nodes' respective resources (e.g., memory, disks, processors). As disclosed herein, embodiments of a computing system hosting an in-memory database may distribute and store data records of the database among one or more nodes. In some embodiments, these nodes are formed into "clusters" of nodes. In some embodiments, these clusters of nodes store portions, or "collections,” of database information.
  • Various embodiments of the system of FIG. 6 provide a computer system executing a feature disambiguation technique that employs an evolving and efficiently linkable feature knowledge base that is configured to store secondary features, such as co- occurring topics, key phrases, proximity terms, events, facts and a trending popularity index.
  • the disclosed embodiments may be performed via various linking algorithms that can vary from simple conceptual distance measure to sophisticated graph clustering approaches based on the dimensions of the involved secondary features that aid in resolving a given extracted feature to a stored feature in the knowledge base.
  • FIG. 7 is a central server computer system 700 for building a knowledge base
  • Document corpus 702 may be any collection of documents such as, for example, a database of digital documents from a company or the World Wide Web.
  • the process may involve crawling through each document in document corpus
  • a document may include information in unstructured text format which may be crawled using natural language processing techniques (NLP).
  • NLP techniques include, for example, removing stop words, tokenization, stemming and part-of speech tagging among others know in the art.
  • An individual file may first go through an entity extraction module 704 where entities (e.g., a person, location, or organization name) are identified and extracted.
  • Entity extraction module 704 may be a software module with programmatic logic that may extract entities by crawling through the document. The extracted entities may then be passed to a knowledge base aggregator 706.
  • the file may then go through a topic extractor module 708, which may be executed by an in-memory database computer.
  • the in-memory database computer can be a database storing data in records controlled by a database management system (DBMS) (not shown) configured to store data records in a device's main memory, as opposed to conventional databases and DBMS modules that store data in "disk" memory.
  • DBMS database management system
  • CPUs processors
  • In-memory database systems access data that is placed into main memory, and then addressed accordingly, thereby mitigating the number of instructions performed by the CPUs and eliminating the seek time associated with CPUs seeking data on hard disk.
  • In-memory databases may be implemented in a distributed computing architecture, which may be a computing system comprising one or more nodes configured to aggregate the nodes' respective resources (e.g., memory, disks, processors).
  • a computing system hosting an in-memory database may distribute and store data records of the database among one or more nodes.
  • these nodes are formed into "clusters" of nodes.
  • these clusters of nodes store portions, or "collections,” of database information.
  • Various embodiments provide a computer executed feature disambiguation technique that employs an evolving and efficiently linkable feature knowledge base that is configured to store secondary features, such as co-occurring topics, key phrases, proximity terms, events, facts and trending popularity index.
  • the disclosed embodiments may be performed via a wide variety of linking algorithms that can vary from simple conceptual distance measure to sophisticated graph clustering approaches based on the dimensions of the involved secondary features that aid in resolving a given extracted feature to a stored feature in the knowledge base.
  • embodiments can introduce an approach to evolves the existing feature knowledge base by a capability that not only updates the secondary features of the existing feature entry, but also expands it by discovering new features that can be appended to the knowledge base.
  • Topic extractor module 708 may extract the theme or topic of a single document file. In most cases a file may include a single topic, however a plurality of topics may also exist in a single document. Topic extraction techniques may include, for example, comparing keywords against latent Dirichlet allocation (LDA) models or any other techniques for topic identification. The extracted topic may also be passed to knowledge base aggregator 706 for further processing.
  • LDA latent Dirichlet allocation
  • System 700 may also include an event detection module 710 for extracting events from a document file.
  • events may include an accident (e.g., car accident, a train accident, etc.), a natural disaster (e.g., an earthquake, a flood, a weather event, etc.), a man-made disaster (e.g., a bridge collapse, a discharge of a hazardous material, an explosion, etc.), a security event (e.g., a terrorist attack, an act of war, etc.), and/or any other event that may trigger a response from authorities and/or first responders and/or may trigger a notification to a large quantity (e.g., greater than some threshold) of user devices (e.g., acts associated with a major sporting event or concert, election day coordination, traffic management due to road construction, etc.).
  • a large quantity e.g., greater than some threshold
  • Event detection module 710 may be a software module with programmatic logic that may detect events by extracting keywords from a file and comparing them against event template models stored in an event concept store database. The extracted events may also be passed to knowledge base aggregator 706 for further processing. [0113] A fact extractor module 712 may also be implemented. Fact extractor module
  • Fact extractor module 712 may be a software module with programmatic logic that may extract facts by crawling through the document.
  • Fact extractor module 712 may extract facts by comparing factual text descriptions in a document and comparing them against a fact-word table. Other methods for identifying facts in documents may also be implemented. Identified facts may be passed to knowledge base aggregator 706 for further processing.
  • Knowledge base aggregator 706 may include a cooccurrence module 714 and a co-occurrence store aggregator 716.
  • Co-occurrence module 714 may be a software module with programmatic logic that may aggregate co-occurrence of features across a plurality of documents and record the count of co-occurrences in cooccurrence store aggregation 716. Whenever the co-occurrence of features across documents in a document corpus 702 exceed a determined threshold the co-occurrence of entities may be added to knowledge base 720 along with any metadata pertaining to the co-occurring features.
  • Metadata that may be added to knowledge base 720 may include, for example, the type of the features, the document from where the co-occurrence was extracted, the document corpus, distance in text between co-occurring features, a confidence score that may serve has an indication that the co-occurrence of the features may resemble unique individual features.
  • a confidence score may be calculated by using parameters such as, for example, number of co-occurrences in a single file, number of co-occurrences in a document corpus, size of document corpus, number of co-occurrences in different document corpora, distance in text from co-occurring features, human verification and or any combination thereof.
  • FIG. 7 illustrates an exemplary embodiment and is in no way limiting the scope of the invention. Additional modules for extracting different features not illustrated in FIG. 7 may also been included and are to be considered within the scope of the invention. All software modules may be implemented in a single computer or in a distributed computer architecture across a plurality of computers, whereby the one or more modules may be embodied on at least one computer readable medium and executed by at least one processor.
  • FIG. 8 is an example embodiment of a co-occurring aggregation method 800 using the system described in FIG. 7.
  • document corpus 802 may include three different document files.
  • Features extracted from first document 804 may include “Bill”, “Gates”, “Microsoft”, “Billionaire”.
  • Features extracted from second document 806 may include “Bill”, “Gates”, “President”, “Microsoft”.
  • Features extracted from third document 808 may include "Melinda” and "Gates”.
  • Co-occurrence module 814 may then crawl each document in document corpus 802, store all possible co-occurring feature combinations for a single document and aggregate them with same feature co-occurrences from the other documents of the same corpus.
  • the aggregation process may be done and stored in co-occurrence store aggregator 816.
  • co-occurrence store aggregator 816 For example in FIG. 8 the entities “Bill” and “Gates” co-occur twice, once in first document 804 and once in third document 808 while the entities "Melinda” and “Gates” co-occur once in second document 806.
  • the exemplary search method for discovering and exploring feature knowledge is applied.
  • a user initiates a search with the name of a feature, the results return six different disambiguated features with the same name.
  • the user decides to narrow the search and indicates in the user interface that a higher threshold or different features for the disambiguation should be used, all the data is processed again in one or more analytics agents and the new set of results returns only two different disambiguated features with the same name.
  • a number of different interfaces serving different purposes for different groups of people are fed from the same MEMDB. Each interface was developed to facilitate the manipulation of the analytical parameters relevant to each application.
  • the exemplary search method for discovering and exploring feature knowledge is applied to images.
  • image processing techniques are utilized to extract features from the documents and suitable analytics modules used to process the search results.
  • a process may correspond to a method, a function, a procedure, a subroutine, a subprogram, etc.
  • a process corresponds to a function
  • its termination may correspond to a return of the function to the calling function or the main function.
  • the various illustrative logical blocks, modules, circuits, and algorithm steps described in connection with the embodiments disclosed herein may be implemented as electronic hardware, computer software, or combinations of both. To clearly illustrate this interchangeability of hardware and software, various illustrative components, blocks, modules, circuits, and steps have been described above generally in terms of their functionality. Whether such functionality is implemented as hardware or software depends upon the particular application and design constraints imposed on the overall system. Skilled artisans may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the present invention.
  • Embodiments implemented in computer software may be implemented in software, firmware, middleware, microcode, hardware description languages, or any combination thereof.
  • a code segment or machine-executable instructions may represent a procedure, a function, a subprogram, a program, a routine, a subroutine, a module, a software package, a class, or any combination of instructions, data structures, or program statements.
  • a code segment may be coupled to another code segment or a hardware circuit by passing and/or receiving information, data, arguments, parameters, or memory contents.
  • Information, arguments, parameters, data, etc. may be passed, forwarded, or transmitted via any suitable means including memory sharing, message passing, token passing, network transmission, etc.
  • a non-transitory processor-readable storage media may be any available media that may be accessed by a computer.
  • such non-transitory processor-readable media may comprise RAM, ROM, EEPROM, CD-ROM or other optical disk storage, magnetic disk storage or other magnetic storage devices, or any other tangible storage medium that may be used to store desired program code in the form of instructions or data structures and that may be accessed by a computer or processor.
  • Disk and disc include compact disc (CD), laser disc, optical disc, digital versatile disc (DVD), floppy disk, and blu-ray disc where disks usually reproduce data magnetically, while discs reproduce data optically with lasers. Combinations of the above should also be included within the scope of computer-readable media.
  • the operations of a method or algorithm may reside as one or any combination or set of codes and/or instructions on a non- transitory processor-readable medium and/or computer-readable medium, which may be incorporated into a computer program product.
  • the various components of the technology can be located at distant portions of a distributed network and/or the Internet, or within a dedicated secure, unsecured and/or encrypted system.
  • the components of the system can be combined into one or more devices or co-located on a particular node of a distributed network, such as a telecommunications network.
  • the components of the system can be arranged at any location within a distributed network without affecting the operation of the system.
  • the components could be embedded in a dedicated machine.
  • the various links connecting the elements can be wired or wireless links, or any combination thereof, or any other known or later developed element(s) that is capable of supplying and/or communicating data to and from the connected elements.
  • module as used herein can refer to any known or later developed hardware, software, firmware, or combination thereof that is capable of performing the functionality associated with that element.
  • determine, calculate and compute, and variations thereof, as used herein are used interchangeably and include any type of methodology, process, mathematical operation or technique.

Landscapes

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

Abstract

L'invention concerne des systèmes et des procédés d'extraction de faits à partir de fichiers texte non structurés. Des modes de réalisation de l'invention consistent à recevoir des fichiers texte en entrée et à effectuer l'extraction et la désambiguïsation d'entités, de sujets et de faits. Les faits sont extraits par comparaison de caractéristiques, telles que des mots clés, avec des modèles de faits, et par association de faits à des événements ou à des sujets. Les faits extraits sont stockés dans un magasin de données. L'invention concerne également des procédés et des systèmes de découverte de "connaissances" dans des corpus stockés, ces procédés et systèmes consistant à appliquer une analyse en mémoire à des enregistrements de base de données en fonction d'une indication sélectionnée par l'utilisateur. L'invention concerne également des systèmes et des procédés de construction d'une base de connaissances au moyen de caractéristiques co-occurentes, telles que des mots clés, extraites de corpus. Des modes de réalisation concernent un logiciel d'extraction de caractéristiques qui extrait des caractéristiques de fichiers documents dans un corpus stocké. Des modes de réalisation peuvent également concerner un module logiciel agrégateur de base de connaissances qui compte le nombre de co-occurences de caractéristiques dans les divers documents d'un corpus et qui identifie les co-occurences de caractéristiques à stocker dans une base de connaissances.
PCT/US2014/067994 2013-12-02 2014-12-02 Systèmes et procédés de traitement de données stockées dans une base de données WO2015084757A1 (fr)

Applications Claiming Priority (6)

Application Number Priority Date Filing Date Title
US201361910880P 2013-12-02 2013-12-02
US201361910883P 2013-12-02 2013-12-02
US201361910887P 2013-12-02 2013-12-02
US61/910,880 2013-12-02
US61/910,883 2013-12-02
US61/910,887 2013-12-02

Publications (1)

Publication Number Publication Date
WO2015084757A1 true WO2015084757A1 (fr) 2015-06-11

Family

ID=53274013

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/US2014/067994 WO2015084757A1 (fr) 2013-12-02 2014-12-02 Systèmes et procédés de traitement de données stockées dans une base de données

Country Status (1)

Country Link
WO (1) WO2015084757A1 (fr)

Cited By (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
EP3133504A3 (fr) * 2015-08-20 2017-04-05 Xiaomi Inc. Procédé et dispositif de construction de base de connaissances
EP3168791A1 (fr) * 2015-11-10 2017-05-17 Fujitsu Limited Procédé et système de validation de données dans un appareil d'extraction de connaissances
EP3223179A1 (fr) * 2016-03-24 2017-09-27 Fujitsu Limited Système et procédé d'extraction de risque en matière de soins de santé
WO2018026489A1 (fr) * 2016-08-05 2018-02-08 Google Llc Émergence de faits uniques pour des entités
US10394555B1 (en) 2018-12-17 2019-08-27 Bakhtgerey Sinchev Computing network architecture for reducing a computing operation time and memory usage associated with determining, from a set of data elements, a subset of at least two data elements, associated with a target computing operation result
US11546142B1 (en) 2021-12-22 2023-01-03 Bakhtgerey Sinchev Cryptography key generation method for encryption and decryption

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20020031260A1 (en) * 2000-06-29 2002-03-14 Ssr Co., Ltd. And Kochi University Of Technology Text mining method and apparatus for extracting features of documents
US20040243645A1 (en) * 2003-05-30 2004-12-02 International Business Machines Corporation System, method and computer program product for performing unstructured information management and automatic text analysis, and providing multiple document views derived from different document tokenizations
US20070156748A1 (en) * 2005-12-21 2007-07-05 Ossama Emam Method and System for Automatically Generating Multilingual Electronic Content from Unstructured Data
US20080077570A1 (en) * 2004-10-25 2008-03-27 Infovell, Inc. Full Text Query and Search Systems and Method of Use
US20090222395A1 (en) * 2007-12-21 2009-09-03 Marc Light Systems, methods, and software for entity extraction and resolution coupled with event and relationship extraction
US20110161333A1 (en) * 2005-07-05 2011-06-30 Justin Langseth System and method of making unstructured data available to structured data analysis tools

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20020031260A1 (en) * 2000-06-29 2002-03-14 Ssr Co., Ltd. And Kochi University Of Technology Text mining method and apparatus for extracting features of documents
US20040243645A1 (en) * 2003-05-30 2004-12-02 International Business Machines Corporation System, method and computer program product for performing unstructured information management and automatic text analysis, and providing multiple document views derived from different document tokenizations
US20080077570A1 (en) * 2004-10-25 2008-03-27 Infovell, Inc. Full Text Query and Search Systems and Method of Use
US20110161333A1 (en) * 2005-07-05 2011-06-30 Justin Langseth System and method of making unstructured data available to structured data analysis tools
US20070156748A1 (en) * 2005-12-21 2007-07-05 Ossama Emam Method and System for Automatically Generating Multilingual Electronic Content from Unstructured Data
US20090222395A1 (en) * 2007-12-21 2009-09-03 Marc Light Systems, methods, and software for entity extraction and resolution coupled with event and relationship extraction

Cited By (12)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
EP3133504A3 (fr) * 2015-08-20 2017-04-05 Xiaomi Inc. Procédé et dispositif de construction de base de connaissances
JP2017532704A (ja) * 2015-08-20 2017-11-02 シャオミ・インコーポレイテッド 知識ベースの構築方法及び装置
US10331648B2 (en) 2015-08-20 2019-06-25 Xiaomi Inc. Method, device and medium for knowledge base construction
EP3168791A1 (fr) * 2015-11-10 2017-05-17 Fujitsu Limited Procédé et système de validation de données dans un appareil d'extraction de connaissances
EP3223179A1 (fr) * 2016-03-24 2017-09-27 Fujitsu Limited Système et procédé d'extraction de risque en matière de soins de santé
WO2018026489A1 (fr) * 2016-08-05 2018-02-08 Google Llc Émergence de faits uniques pour des entités
CN107688616A (zh) * 2016-08-05 2018-02-13 谷歌有限责任公司 使实体的独特事实显现
CN107688616B (zh) * 2016-08-05 2021-07-09 谷歌有限责任公司 使实体的独特事实显现
US11568274B2 (en) 2016-08-05 2023-01-31 Google Llc Surfacing unique facts for entities
US10394555B1 (en) 2018-12-17 2019-08-27 Bakhtgerey Sinchev Computing network architecture for reducing a computing operation time and memory usage associated with determining, from a set of data elements, a subset of at least two data elements, associated with a target computing operation result
US10860317B2 (en) 2018-12-17 2020-12-08 Bakhtgerey Sinchev Computing network architecture for reducing computing operation time, memory usage, or other computing resource usage, associated with determining, from a set of data elements, at least two data elements, associated with a target computing operation result
US11546142B1 (en) 2021-12-22 2023-01-03 Bakhtgerey Sinchev Cryptography key generation method for encryption and decryption

Similar Documents

Publication Publication Date Title
US9424524B2 (en) Extracting facts from unstructured text
US9239875B2 (en) Method for disambiguated features in unstructured text
Bharti et al. Sarcastic sentiment detection in tweets streamed in real time: a big data approach
US9922032B2 (en) Featured co-occurrence knowledge base from a corpus of documents
US9720944B2 (en) Method for facet searching and search suggestions
Ramnandan et al. Assigning semantic labels to data sources
US9619571B2 (en) Method for searching related entities through entity co-occurrence
US9201931B2 (en) Method for obtaining search suggestions from fuzzy score matching and population frequencies
US20170116203A1 (en) Method of automated discovery of topic relatedness
US20170286837A1 (en) Method of automated discovery of new topics
US20160048754A1 (en) Classifying resources using a deep network
US20130060769A1 (en) System and method for identifying social media interactions
KR20180011254A (ko) 웹페이지 트레이닝 방법 및 기기, 그리고 검색 의도 식별 방법 및 기기
Reinanda et al. Mining, ranking and recommending entity aspects
WO2015084757A1 (fr) Systèmes et procédés de traitement de données stockées dans une base de données
US20170109358A1 (en) Method and system of determining enterprise content specific taxonomies and surrogate tags
Tayal et al. Fast retrieval approach of sentimental analysis with implementation of bloom filter on Hadoop
CN103226601A (zh) 一种图片搜索的方法和装置
Wei et al. Online education recommendation model based on user behavior data analysis
US20170124090A1 (en) Method of discovering and exploring feature knowledge
US9223833B2 (en) Method for in-loop human validation of disambiguated features
US20160246794A1 (en) Method for entity-driven alerts based on disambiguated features
SCALIA Network-based content geolocation on social media for emergency management
KR102625347B1 (ko) 동사와 형용사와 같은 품사를 이용한 음식 메뉴 명사 추출 방법과 이를 이용하여 음식 사전을 업데이트하는 방법 및 이를 위한 시스템
Otsuka et al. Text Filtering for Harmful Document Classification Using Three‐Word Co‐Occurrence and Large‐Scale Data Processing

Legal Events

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

Ref document number: 14867620

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 14867620

Country of ref document: EP

Kind code of ref document: A1