WO2019177182A1 - Appareil de recherche de contenu multimédia et procédé de recherche utilisant une analyse d'informations d'attributs - Google Patents

Appareil de recherche de contenu multimédia et procédé de recherche utilisant une analyse d'informations d'attributs Download PDF

Info

Publication number
WO2019177182A1
WO2019177182A1 PCT/KR2018/002911 KR2018002911W WO2019177182A1 WO 2019177182 A1 WO2019177182 A1 WO 2019177182A1 KR 2018002911 W KR2018002911 W KR 2018002911W WO 2019177182 A1 WO2019177182 A1 WO 2019177182A1
Authority
WO
WIPO (PCT)
Prior art keywords
search
attribute
information
unit
multimedia content
Prior art date
Application number
PCT/KR2018/002911
Other languages
English (en)
Korean (ko)
Inventor
송민규
Original Assignee
미디어젠 주식회사
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 미디어젠 주식회사 filed Critical 미디어젠 주식회사
Publication of WO2019177182A1 publication Critical patent/WO2019177182A1/fr

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/40—Information retrieval; Database structures therefor; File system structures therefor of multimedia data, e.g. slideshows comprising image and additional audio data
    • G06F16/43—Querying
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/40—Information retrieval; Database structures therefor; File system structures therefor of multimedia data, e.g. slideshows comprising image and additional audio data
    • G06F16/43—Querying
    • G06F16/432—Query formulation
    • G06F16/433—Query formulation using audio data
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/40—Information retrieval; Database structures therefor; File system structures therefor of multimedia data, e.g. slideshows comprising image and additional audio data
    • G06F16/48—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually

Definitions

  • the present invention relates to an apparatus and method for searching multimedia contents through attribute information analysis. More particularly, the present invention relates to a method of searching a text keyword by acquiring a search word of multimedia content input by speech recognition or text, or performing a similar attribute search. The multimedia information to be searched is output by outputting the search result information of the multimedia contents when performing the text keyword search by determining whether to perform the search.
  • the present invention relates to a multimedia content retrieval apparatus and a retrieval method through attribute information analysis capable of providing multimedia contents having high similarity with the attribute information of the content.
  • a portal company such as the following, and a search engine such as Google
  • the user can search for the latest keyword information related to the keyword of the user's search query, or a specific operator grouping the keywords. Through this, efforts are made to provide information closer to the information desired by the user.
  • the related search word providing service not only facilitates a user's search, but also serves as one piece of information.
  • the prior art 1 relates to a keyword visualization apparatus and a method thereof, comprising: a keyword extracting unit extracting a keyword from data obtained through the Internet; A frequency analysis unit for raising a frequency of occurrence of the keyword each time a keyword is extracted; An association analysis unit for increasing association values between the extracted keywords when a plurality of keywords are extracted from a single data; An information storage unit for storing the extracted keywords and storing occurrence frequency values for each keyword and correlation values between the keywords; And a graph having a plurality of nodes and edges is displayed on the screen by using a plurality of keywords, occurrence frequency values of the keywords, and correlation values between the keywords, and each node of the graph is displayed with keywords.
  • Nodes with high values are displayed in large sizes, and nodes with low keyword occurrence frequencies are displayed in small sizes. If the correlation values between keywords of two nodes connected by edges are high, the edges are displayed with thick edges. If it is low, characterized in that it comprises a visualization processing unit for processing so that the edge is displayed thin, suggests a change in the frequency of occurrence of the keyword and the degree of association between the keywords.
  • patent documents include "a search method and system using the ranking of keywords (patent registration no. 10-1072113, hereinafter referred to as” prior art 2 ").
  • the prior art 2 is a search method and system using an association ranking of a keyword, comprising: an index module for generating an independent index by indexing a property of a keyword and an association index by indexing a correlation between a keyword and another keyword; An association score calculation module that quantifies an association degree between a keyword and another keyword based on an association index as an association score; A rank score calculation module that calculates a rank score according to the use purpose based on the association score and the independent index; And a search module for providing a related keyword for the search term based on the ranking score.
  • Prior Art 2 only discloses a technical idea of extracting a related search word for a keyword, and does not provide general information on the related search.
  • Prior Art 1 provides a graph of ranking among related search terms for a keyword to provide which related search terms for a search term is the most frequently used.
  • the related art automatically searches for the highest frequency among related search terms. It is not much different from the known technology ranking at the top of the related search word list.
  • search systems equipped with artificial intelligence-based can be divided into crawler-based, directory-based, hybrid search, and meta-search method in terms of search method.
  • the crawler-based retrieval system downloads and stores documents on the web in its database using an automated agent program called spider, crawler, webbot, and the like.
  • the user's search request is handled by finding the search keyword in the index of the stored web document and providing a link to that document.
  • web sites are classified and stored in a predetermined directory by a person, and the stored websites are ranked by a predefined rule.
  • the user's search request is processed by grouping the web documents found by keyword matching by directory.
  • the crawler method and the directory method are used together and generally provide a better search result to the user.
  • meta-search system utilizes search algorithms and evaluation criteria of other search systems.
  • search results of different search systems are merged and provided to the user.
  • Metacrawler system is a typical example.
  • a first object of the present invention is to perform a text keyword search by acquiring a search word of multimedia content input by speech recognition or text, or similar property search.
  • the search result information of the multimedia content is output when the text keyword search is performed and the search result information of the multimedia content having the similar property is output.
  • the second object of the present invention is to provide the similarity matching property analysis unit 530 to the attribute information and the multimedia content attribute assignment unit 520 for the search word stored in the search term attribute value information DB 517 when performing the similar attribute search. Similarity matching analysis is performed with the multimedia content attribute information assigned by the present invention, thereby providing search result information of the multimedia content having attributes similar to the intention of the search word (question).
  • the third object of the present invention is to provide a content crawling module 522, to collect a plurality of multimedia content information from the content server 560 to store in the content storage DB to extend the operation range of the attribute information, the content attribute allocation model
  • the module 524 By providing the module 524, the attribute information is allocated to each multimedia content stored in the content storage DB 523 and provided to the content information search module.
  • the multimedia content retrieval apparatus through attribute information analysis
  • a search start unit 100 for acquiring a search word of multimedia content input by voice recognition or text and providing search execution request information to the attribute search execution determining unit 200;
  • the search execution request information from the search start unit 100 it is determined whether to perform a text keyword search or a similar attribute search, and as a result of the determination, the text keyword search is performed when the text keyword search is performed.
  • the attribute search decision unit 200 which provides the text keyword search request information to the unit 300, and provides the similar property search request information to the attribute similarity search unit 500 when performing the similar attribute search as a result of the determination; ,
  • a text keyword search unit 300 which performs a text keyword search when obtaining the text keyword search request information provided from the attribute search performing determination unit, and provides the search result information to the text keyword result output unit;
  • a text keyword result output unit 400 for outputting search result information of the text keyword provided from the text keyword search unit;
  • An attribute similarity search means 500 which performs a similar attribute search when obtaining similar attribute search request information provided from the attribute search execution determination unit and provides the search result information to the attribute similarity search result output unit 500;
  • an attribute similarity search result output unit 600 for outputting search result information of the similar attribute provided from the attribute similarity search unit 500.
  • the multimedia content retrieval method by analyzing the attribute information
  • the attribute search execution unit 200 obtains the search execution request information from the search start unit 100, it is determined whether to perform a text keyword search or a similar attribute search.
  • the text keyword search request information is provided to the text keyword search unit 300, and as a result of the determination, when the similar property search is performed, the similar property search request information is provided to the attribute similarity search unit 500.
  • Attribute search determination step (S200) is performed when the attribute search execution request information is performed.
  • the text keyword search unit 300 obtains the text keyword search request information provided from the attribution search execution determination unit 200, the text keyword search unit performs a text keyword search and provides the search result information to the text keyword result output unit. Step S300,
  • the attribute similarity search result output unit 600 includes an attribute similarity search result output step S600 for outputting search result information of similar attributes provided from the attribute similarity search unit 500.
  • Search result of multimedia content is output when performing a text keyword search by determining whether to perform a text keyword search, and output search result information of multimedia content having a similar property when performing a similar property search.
  • the present invention provides an effect of providing a multimedia content search result using a keyword method and of providing a multimedia content search result most similar to a search word (question) that a user wants to search through a similar property search.
  • the amount of information of the multimedia content changes over time, and accordingly, the attributes of a specific object change from time to time.
  • the multimedia content attribute assignment unit By reflecting this variably by the multimedia content attribute assignment unit, various multimedia contents that change in real time may be reflected in a search. Will be effective.
  • FIG. 1 is an overall configuration diagram schematically showing an apparatus for retrieving multimedia contents through attribute information analysis according to a first embodiment of the present invention.
  • FIG. 2 is an exemplary view in which a movie of a conventional similar atmosphere is not searched.
  • FIG. 3 is an overall block diagram of an apparatus for retrieving multimedia contents through attribute information analysis according to a first embodiment of the present invention.
  • FIG. 4 is an exemplary view showing a search result output when a text keyword is searched.
  • FIG. 5 is an exemplary view of a similar property search result output through a multimedia content search apparatus through analysis of property information according to a first embodiment of the present invention.
  • FIG. 6 is a block diagram of attribute similarity retrieval means of a multimedia content retrieval apparatus by analyzing attribute information according to the first embodiment of the present invention
  • FIG. 7 is a block diagram of a keyword attribute analysis unit of a multimedia content retrieval apparatus through attribute information analysis according to the first embodiment of the present invention.
  • FIG. 8 is a block diagram of a multimedia content attribute assignment unit of the multimedia content retrieval apparatus through attribute information analysis according to the first embodiment of the present invention.
  • FIG. 9 is a flowchart illustrating a multimedia content retrieval method through attribute information analysis according to a first embodiment of the present invention.
  • FIG. 10 is a flowchart illustrating an attribute similarity search step of a multimedia content search method through analysis of attribute information according to a first embodiment of the present invention
  • first and second may be used to describe various components, but the components may not be limited by the terms.
  • the first component may be referred to as the second component, and similarly, the second component may also be referred to as the first component.
  • a component When a component is referred to as being connected or connected to another component, it may be understood that the component may be directly connected to or connected to the other component, but there may be other components in between. .
  • an apparatus for retrieving multimedia contents through attribute information analysis In accordance with a first aspect of the present invention, there is provided an apparatus for retrieving multimedia contents through attribute information analysis.
  • a search start unit 100 for acquiring a search word of multimedia content input by voice recognition or text and providing search execution request information to the attribute search execution determining unit 200;
  • the search execution request information from the search start unit 100 it is determined whether to perform a text keyword search or a similar attribute search, and as a result of the determination, the text keyword search is performed when the text keyword search is performed.
  • the attribute search decision unit 200 which provides the text keyword search request information to the unit 300, and provides the similar property search request information to the attribute similarity search unit 500 when performing the similar attribute search as a result of the determination; ,
  • a text keyword search unit 300 which performs a text keyword search when obtaining the text keyword search request information provided from the attribute search performing determination unit, and provides the search result information to the text keyword result output unit;
  • a text keyword result output unit 400 for outputting search result information of the text keyword provided from the text keyword search unit;
  • An attribute similarity search means 500 which performs a similar attribute search when obtaining similar attribute search request information provided from the attribute search execution determination unit and provides the search result information to the attribute similarity search result output unit 500;
  • It is characterized in that it comprises a property similarity search result output unit 600 for outputting the search result information of the similar property provided from the attribute similarity search unit 500.
  • a search word attribute analyzer 510 for analyzing linguistic attribute information included in a search word of multimedia content input through speech recognition or text;
  • a multimedia content attribute allocator 520 for acquiring and storing multimedia contents from the content server 560 and allocating attribute information to the stored multimedia contents;
  • a similarity matching analysis unit 530 for performing a similarity matching analysis of multimedia contents included in the multimedia contents list information
  • a similarity candidate group extracting unit 540 for sequentially extracting multimedia contents according to candidate group numbers from multimedia contents having the highest similarity with reference to a preset candidate group number;
  • a similarity reference multimedia content sorting unit 550 for sorting the multimedia contents extracted according to the number of candidate groups according to similarity and providing the sorted multimedia contents to the attribute similarity search result output unit 600. do.
  • the machine learning model module 512 provides information on requesting interpretation of linguistic attributes included in a search word of multimedia content input through speech recognition or text, and provides linguistic attribute information included in a search word interpreted from the machine learning model module.
  • Machine learning model module for providing linguistic attribute information interpreted as natural language processing module by interpreting linguistic attributes included in search term when obtaining information on interpretation of linguistic attributes included in search term from natural language processing module. 512);
  • a knowledge information DB 514 that stores attribute type information refined into attribute types that can be matched with attribute information of multimedia content
  • the probability model calculation request information is provided to the attribute model module 516, and the probability value calculated from the attribute model module 516 is obtained to provide the search term.
  • a keyword attribute value conversion module 515 for converting the attribute value into an attribute value and providing the result to the keyword attribute value information DB 517;
  • An attribute model module 516 for calculating a probability value through language modeling when obtaining the probability value calculation request information from the keyword attribute value conversion module 515 and providing the calculated probability value to the keyword attribute value conversion module 515;
  • a search word attribute value information DB 517 that stores the attribute value for the search word provided by the search word attribute value conversion module 515.
  • a content interlocking module 521 for providing multimedia content information to the content crawling module 522 in association with the content server 560;
  • a content crawling module 522 for collecting a plurality of multimedia content information provided from the content interworking module 521 and storing the multimedia content information in a content storage DB to expand the operation range of the attribute information;
  • a content storage DB 523 for storing multimedia content information provided from the content crawling module 522 and attribute information allocated to each multimedia content
  • a content property information analysis module 525 for analyzing the property information of each multimedia content assigned by the content property assignment model module 524 and providing the same to the content information search module;
  • the attribute information of each multimedia content analyzed by the content attribute information analysis module 525 is provided to the similarity matching property analysis unit 530, and similar property information is similar to the linguistic property information of the search word from the similarity matching property analysis unit 530.
  • the similarity matching analysis may be performed using the attribute information of the search word stored in the search term attribute value information DB 517 and the multimedia content attribute information allocated by the multimedia content attribute assigning unit 520.
  • a method for retrieving multimedia contents by analyzing attribute information includes:
  • the attribute search execution unit 200 obtains the search execution request information from the search start unit 100, it is determined whether to perform a text keyword search or a similar attribute search.
  • the text keyword search request information is provided to the text keyword search unit 300, and as a result of the determination, when the similar property search is performed, the similar property search request information is provided to the attribute similarity search unit 500.
  • Attribute search determination step (S200) is performed when the attribute search execution request information is performed.
  • the text keyword search unit 300 obtains the text keyword search request information provided from the attribution search execution determination unit 200, the text keyword search unit performs a text keyword search and provides the search result information to the text keyword result output unit. Step S300,
  • the attribute similarity search result output unit 600 includes an attribute similarity search result output step S600 for outputting search result information of similar attributes provided from the attribute similarity search unit 500.
  • a multimedia content attribute assignment step (S520) of the multimedia content attribute assignment unit 520 acquiring and storing multimedia content from the content server 560 and allocating attribute information to the stored multimedia content;
  • the similarity matching property analysis unit 530 provides the multimedia content property assignment unit 520 with multimedia content request information including property information similar to the linguistic property information of the search word, and the multimedia content from the multimedia content property assignment unit 520.
  • Similarity-based multimedia content sorting unit 550 sorts the multimedia contents extracted according to the number of candidate groups according to similarity, and provides similarity-based multimedia content sorting step to provide the sorted multimedia contents to the attribute similarity search result output unit 600.
  • S550 characterized in that it comprises a.
  • FIG. 1 is an overall configuration diagram schematically showing an apparatus for retrieving multimedia contents through attribute information analysis according to a first embodiment of the present invention.
  • the apparatus 1000 for retrieving multimedia contents through the analysis of attribute information of the present invention obtains and stores multimedia contents from the content server 560, and allocates and manages attribute information to the stored multimedia contents. to be.
  • the multimedia content search apparatus 1000 through attribute information analysis acquires a search word of multimedia content input by voice recognition or text, and determines whether to perform a text keyword search or a similar property search.
  • the search result information of the multimedia content is output.
  • the conventional text keyword based search has a problem of being searched again with the same title, and a movie having a similar name and a completely different content is recommended. There was a serious problem that the movies were not searched at all.
  • the user cannot search for a movie that has a similar mood, emotion, or the like.
  • the present invention by providing the above-described text keyword-based search function, by providing a structural feature for performing a similar property search, the search results of multimedia content having similar properties when performing a similar property search By outputting the information, it is possible to provide multimedia contents having high similarity to the attribute information of the multimedia contents to be searched.
  • the present invention through the configuration as described above, to determine whether to proceed to the existing keyword search or similar property search and the attributes of each multimedia content (warm, touching, fun, etc.)
  • the similarity between the constructive feature that assigns the attribute value of the searched multimedia through the constructive feature and natural language processing (data crawling, statistical modeling, etc.) and the comparable feature and attribute information that are numerically calculated (language modeling) It provides a constructive feature for recommending high multimedia content (comparison value).
  • FIG. 3 is a block diagram of an apparatus for retrieving multimedia contents through attribute information analysis according to a first embodiment of the present invention.
  • the present invention provides a multimedia content search apparatus 1000 through attribute information analysis.
  • the search start unit 100, the attribute search execution determination unit 200, the text keyword search unit 300, and the text keyword result are shown. It comprises an output unit 400, the attribute similarity search means 500, the attribute similarity search result output unit 600.
  • the present invention provides a text keyword type search and an attribute similarity type search.
  • the search start unit 100 obtains a search word of multimedia content input through voice recognition or text and provides search execution request information to the attribute search execution determining unit 200.
  • the search start unit includes a natural language processing module for speech recognition, and extracts a user's command target value from the speech recognition result text processed by the natural language processing module.
  • Embedded Natural Language Understanding technology incorporates a natural language processing module using a rule-based algorithm or statistical model inside an electronic device, so that the user's final goal in speech recognition text is a command. It means the method of automatically extracting the intention (Intention, Goal) and the specific named object, it is to extract the command target value of the user from the speech recognition result text processed by the natural language processing module.
  • the search start unit may configure a voice recognition engine, through which the function of extracting a recognition result value by recognizing a result close to a word or sentence previously input as a command based on the extracted command target value of the user. Done.
  • speech recognition is performed based on recognition grammars that can be understood by a recognizer, and a list of recognition targets is determined, and only the target list has a structure that can be output as a recognition result.
  • the search start unit 100 obtains a search word of the multimedia content input through voice recognition or text and provides the search execution request information to the attribute search execution determining unit 200.
  • a user inputs a movie such as a love act by voice or text, it can be referred to as a search word for requesting multimedia content by referring to a love act, a movie, and the like. It will be provided to the search performance determination unit 200.
  • the attribute search determining unit 200 determines whether to perform a text keyword search or a similar attribute search when obtaining the search execution request information from the search start unit 100.
  • the determination of whether to perform a text keyword search or a similar property search is performed in at least one of a first mode for determining according to a service domain and a second mode for analyzing and determining a sentence input by a search word. It is characterized by applying the mode.
  • the first mode or the second mode may be set in advance by an administrator.
  • the first mode when the first mode is set to determine whether to perform a text keyword search or similar property search, whether to perform a text keyword search with reference to the service domain address or similar property search is performed. Is determined.
  • a text keyword search is set for a domain address of 'www.naver.com'
  • a similar attribute search is set for a domain address of 'www.google.com'.
  • a sentence input as a search word is analyzed to determine whether a keyword corresponding to a similar attribute search exists.
  • a search word that is intended to search for similar attributes, such as 'same', 'similar', 'same', etc., it may be understood that this is to perform a similar attribute search.
  • the attribution search performing decision unit 200 provides the text keyword search request information to the text keyword search unit 300 when the text keyword search is performed.
  • the text keyword search unit 300 when the text keyword search unit 300 obtains the text keyword search request information provided from the attribute search performing determination unit, the text keyword search unit 300 performs a text keyword search by referring to the text keyword 'love actual', The search result information including 'Love Actually' is provided to the text keyword result output unit.
  • the text keyword result output unit 400 outputs search result information of the text keyword provided from the text keyword search unit.
  • the present invention is characterized by providing a similar attribute search method while providing a general text keyword search method.
  • the attribute search performing decision unit 200 provides similar attribute search request information to the attribute similarity search unit 500 when performing the similar attribute search.
  • the similarity property search is performed by the property similarity search unit 500. It is to provide the request information.
  • the attribute similarity search means 500 performs a similar attribute search when obtaining the similar attribute search request information provided from the attribute search determining unit 200, and provides the search result information to the attribute similarity search result output unit.
  • the attribute similarity search result output unit 600 outputs the search result information of the similar attribute provided from the attribute similarity search unit 500.
  • a similar property search is performed through the property similarity search means 500, and the search result is provided to the property similarity search result output unit 600 and displayed on the screen. Will print.
  • the attribute similarity search means 500 includes a keyword attribute analysis unit 510, a multimedia content attribute assignment unit 520, a similarity matching property analysis unit 530, a similarity candidate group extraction unit 540, and similarity degree.
  • the reference multimedia content alignment unit 550 is included.
  • the property refers to an inherent characteristic of the object, and the property itself is not meaningful. However, when an object is composed of related properties, one important expression can be expressed, and the property is generally meaningful data. It is recognized as the smallest logical unit of and used for database processing.
  • the similar property is used to search for multimedia content information having the highest similarity with a search word (question or query word).
  • the keyword attribute analyzer 510 analyzes linguistic attribute information included in a keyword of a multimedia content input through speech recognition or text.
  • a linguistic meaning included in a search word such as a movie such as a love reality is analyzed, which means to analyze linguistic attribute information.
  • attribute information such as 'warmness, inspiration, and fun' is assigned to the love reality, it is possible to search for a movie having the above-mentioned attribute information 'warmness, inspiration and fun'.
  • the multimedia content attribute assignment unit 520 acquires and stores multimedia content from the content server 560 and allocates attribute information to the stored multimedia content.
  • the content information is gathered to determine what attribute information the multimedia contents have.
  • the content information is crawled by a connected content server using an external network or communication, and the attribute information is assigned through linguistic refinement.
  • the similarity matchability analysis unit 530 provides the multimedia content attribute assignment unit 520 with multimedia content request information including attribute information similar to linguistic attribute information of a search word.
  • multimedia attribute information such as 'movie', 'love truth', and 'like', which are the linguistic attribute information of the search word
  • multimedia content attribute information such as 'warm, touching, fun'
  • Multimedia content request information including attribute information similar to “im, fun,” and the like
  • Similarity matching analysis of multimedia contents included in the content list information is performed.
  • the similarity matching analysis described above is content to be provided to the user by using various similarity calculation formulas such as Euclidean distance formula and vector space model, which are frequently used to search for similarity in information retrieval theory. Can be selected.
  • the most similar content with the keyword of the content may be searched and the contents may be sorted in the order of high similarity.
  • the number of contents derived as a result of the similarity search may be determined by sorting an upper predetermined number, and the predetermined number may be arbitrarily set by the user according to a situation.
  • a is a keyword inputted by a user to search for content, and there are n keywords in total up to a 1 , a 2 , a 3 ... a n , and the total n keywords are a (a 1 , a 2 , a 3 ... a n)
  • b is the content
  • the total n keywords are b (b 1 , When b 2 , b 3 ... b n)
  • the Euclidean distance formula can be expressed as follows.
  • vector space model can be expressed as follows.
  • Equation 2 the closer the value derived through Equation 2 is to 1, the higher the similarity, and the closer to 0, the lower the similarity may be determined.
  • the similarity between the search keyword and the keyword generated for each content may be inspected by Equation 1 and Equation 2 to sort the contents in the order of high similarity.
  • the similarity candidate group extracting unit 540 extracts the multimedia contents according to the number of candidate groups sequentially from the multimedia contents having the highest similarity with reference to a preset candidate group number.
  • the multimedia content is sequentially extracted according to the number of candidates, and four candidate groups of 'if only, romantic holiday, notting hill, and work-to-member' are extracted.
  • the similarity-based multimedia content sorting unit 550 sorts the multimedia contents extracted according to the number of candidate groups according to the similarity, and provides the sorted multimedia contents to the attribute similarity search result output unit 600.
  • the Euclidean distance formula is The smaller the similarity value is, the higher the similarity is. Therefore, when the content is rearranged in the order of high similarity, the information is sorted in order of work-to-member, romantic holiday, if only, and notting hill, and the corresponding information is returned to the attribute similarity search result output unit 600. Will be provided to the screen.
  • the search term attribute analysis unit 510 includes a natural language processing module 511, a machine learning model module 512, a search term attribute assignment module 513, a knowledge information DB 514, a search term attribute value conversion module 515, and an attribute model. Module 516, search word attribute value information DB (517).
  • the natural language processing module 511 provides the machine learning model module 512 to provide information on requesting interpretation of linguistic attributes included in a search word of multimedia content input through speech recognition or text, and a search word interpreted from the machine learning model module.
  • the linguistic attribute information included in the search word attribute assignment module 513 is provided.
  • the machine learning model module 512 obtains request information for interpretation of linguistic attributes included in the search word from the natural language processing module, the linguistic language interpreted by the natural language processing module is interpreted. Function to provide attribute information.
  • the linguistic attribute information such as 'Love Actually, Movie,' It is provided to the allocation module (513).
  • the knowledge information DB 514 stores attribute type information refined into a type of attribute that can be matched with attribute information of multimedia content.
  • attribute information such as 'warm, touching, fun, romance' as attribute information of a movie called love act
  • 'movie' as an attribute type that can be matched and stored.
  • the type of attribute may be used to find information, a website, a news / region / shopping, a specific field of content, or a multimedia content.
  • the search word attribute assignment module 513 obtains the linguistic attribute information included in the search word provided by the natural language processing module, extracts the attribute type information from the knowledge information DB based on the obtained linguistic attribute information, and then searches the attribute for the search term. And assign the attribute information on the assigned keyword to the keyword attribute value conversion module 515.
  • the attribute type information 'movie' is extracted from the knowledge information DB, and the attribute information of the search term 'warmness, emotion, fun, romance' 'And the like are provided to the keyword attribute value conversion module 515.
  • the search word attribute value conversion module 515 provides the probability model calculation request information to the attribute model module 516 when obtaining the attribute information for the search word provided from the search word attribute assignment module 513.
  • the attribute model module 516 calculates a probability value through language modeling when obtaining the probability value calculation request information from the keyword attribute value conversion module 515, and converts the calculated probability value to the keyword attribute value conversion module 515. Will be provided.
  • the language modeling refers to an algorithm for finding regularity about a grammar, phrase, word, etc. in a natural language and increasing the accuracy of an object to be searched using the regularity.
  • a commonly used method is a statistical modeling method for calculating a probability value, which is a method of expressing a language rule as a probability in a large corpus and restricting the search area through the probability value.
  • N-Gram which is a statistical language model in most language modeling applications, is known as the most successful language model, and the present invention preferably uses N-Gram.
  • the keyword attribute value conversion module 515 obtains the probability value calculated from the attribute model module 516, converts the probability value into an attribute value for the keyword, and provides the result to the keyword attribute value information DB 517.
  • the attribute information is converted into attribute values for each attribute information and stored in the query attribute value information DB 517.
  • the attribute information for the search word is also stored.
  • the similarity matching property analysis unit 530 has similarity with the attribute values for various search terms and contents provided by the multimedia content attribute assignment unit 520 described below. Will be analyzed.
  • the similarity matching analysis unit 530 performs similarity matching analysis using the attribute information on the search word stored in the search word attribute value information DB 517 and the multimedia content property information allocated by the multimedia content attribute assigning unit 520. will be.
  • the similarity matching analysis unit 530 obtains the multimedia content list information and performs the similarity matching analysis.
  • the multimedia content attribute assignment unit 520 includes a content linkage module 521, a content crawling module 522, a content storage DB 523, a content attribute assignment model module 524, a content attribute information interpretation module 525, and a content. And an information retrieval module 526.
  • the amount of information of multimedia contents changes with the passage of time, and accordingly, the attributes of a specific object change from time to time, and various multimedia contents that are changed in real time are searched by reflecting multimedia contents variably through the multimedia content attribute assignment unit as described above. The effect can be reflected in.
  • the content interlocking module 521 interoperates with the content server 560 to provide the multimedia content information to the content crawling module 522, and the content crawling module 522 is provided from the content interlocking module 521. Collecting a plurality of multimedia content information provided and stored in the content storage DB to extend the operation range of the attribute information.
  • the information delivered from the content server becomes a resource of the content property model through the content interworking module.
  • the multimedia content is collected through the content crawling module 522 to expand the operation range of the attribute information.
  • the content property assignment model module 524 obtains each multimedia content stored in the content storage DB 523 and allocates property information to each multimedia content.
  • the content storage DB 523 stores multimedia content information provided from the content crawling module 522 and attribute information allocated to each multimedia content.
  • it plays a role of assigning attribute information to each multimedia content, for example, assigning attribute information of 'calm and touching' to A music.
  • the content attribute information analysis module 525 interprets the attribute information of each multimedia content assigned by the content attribute assignment model module 524 and provides the same to the content information search module.
  • a content information search module requests a 'movie' that provides 'warmness, inspiration, fun, romance' corresponding to a search word
  • the corresponding content is interpreted, and each of the analyzed multimedia contents is analyzed.
  • the attribute information is provided to the content information search module 526.
  • the content information retrieval module 526 is to provide the similarity matching property analysis unit 530 with attribute information of each multimedia content analyzed by the content property information analysis module 525.
  • the multimedia content request information including attribute information similar to the linguistic attribute information of the search word is obtained from the similarity matching property analysis unit 530, the multimedia content list including the similar attribute information from the content storage DB 523.
  • the information is requested to the content attribute information analysis module 525, the multimedia content list information including similar attribute information is obtained from the content storage DB 523, and provided to the similarity matching property analysis unit 530.
  • the multimedia contents list information such as 'If Only, Romantic Holiday, Notting Hill, Work to Remember' including similar attribute information is stored in the content storage DB. It is extracted from 523.
  • FIG. 9 is a flowchart illustrating a multimedia content searching method through attribute information analysis according to a first embodiment of the present invention.
  • the multimedia content search method through attribute information analysis includes: a search start step (S100), an attribute search execution determination step (S200), a text keyword search step (S300), and a text keyword result output step (S400). ), Attribute similarity search step (S500), and attribute similarity search result output step (S600).
  • the search start step (S100) is to obtain the search request request information of the multimedia content input by voice recognition or text through the search start unit 100 to provide the search execution request information to the attribute search determination unit 200. Done.
  • the search information is provided by extracting text information and providing search request information.
  • the search start unit includes a natural language processing module for voice recognition, and processes the voice processed by the natural language processing module.
  • the command target value of the user is extracted from the recognition result text.
  • the attribute search determination step (S200) when the attribute search execution determination unit 200 obtains the search execution request information from the search start unit 100, whether to perform a text keyword search or perform a similar attribute search If it is determined whether or not to perform, and as a result of the determination, the text keyword search unit 300 provides the text keyword search request information when performing the text keyword search, and when the similar attribute search is performed, the attribute similarity search unit ( 500), similar property search request information is provided.
  • the text keyword search request information is provided to the text keyword search unit 300.
  • the text keyword search step (S300) performs a text keyword search when the text keyword search unit 300 obtains the text keyword search request information provided from the attribution search execution determination unit 200, and retrieves the search result information. It is provided to the text keyword result output unit.
  • the text keyword result output unit 400 outputs search result information of the text keyword provided from the text keyword search unit 300.
  • a text keyword search is performed by referring to a text keyword called 'love actual', and search result information including 'love actual' is provided to the text keyword result output unit.
  • the attribute search decision unit 200 when the attribute search decision unit 200 performs a similar attribute search as a result of the determination, it provides the similarity attribute search request information to the attribute similarity search unit 500, in which the attribute similarity search step (S500)
  • the similarity similarity search means 500 obtains the similar property search request information provided from the attribution search execution decision unit 200, the similar property search is performed and the search result information is provided to the property similarity search result output unit. .
  • the attribute similarity search result output unit 600 outputs search result information of similar attributes provided from the attribute similarity search unit 500.
  • a similar property search is performed through the property similarity search means 500, and the search result is provided to the property similarity search result output unit 600 and displayed on the screen. Will print.
  • FIG. 10 is a flowchart illustrating an attribute similarity retrieval step of a multimedia content retrieval method through attribute information analysis according to a first embodiment of the present invention.
  • the attribute similarity search step (S500), the keyword attribute analysis step (S510), multimedia content attribute assignment step (S520), similarity matching property analysis step (S530), similarity candidate group extraction step (540), Similarity-based multimedia content sorting step (S550) is included.
  • the search word attribute analyzer 510 analyzes linguistic attribute information included in a search word of multimedia content input by voice recognition or text.
  • attribute information such as 'warmness, inspiration, and fun' is assigned to the love reality, it is possible to search for a movie having the above-mentioned attribute information 'warmness, inspiration and fun'.
  • the multimedia content attribute assignment unit 520 acquires and stores the multimedia content from the content server 560, and assigns attribute information to the stored multimedia content.
  • the content information is gathered to determine what attribute information the multimedia contents have.
  • the content information is crawled by a connected content server using an external network or communication, and the attribute information is assigned through linguistic refinement.
  • the similarity matching property analysis unit 530 provides the multimedia content property assignment unit 520 with the multimedia content request information including the similar property information in the linguistic property information of the search word.
  • the multimedia content list information is obtained from the content property allocator 520, and similarity matching analysis of multimedia contents included in the obtained multimedia content list information is performed.
  • multimedia attribute information such as 'movie', 'love truth', and 'like', which are the linguistic attribute information of the search word
  • multimedia content attribute information such as 'warm, touching, fun'
  • Multimedia content request information including attribute information similar to “im, fun,” and the like
  • Similarity matching analysis of multimedia contents included in the content list information is performed.
  • the similarity candidate group extracting unit 540 extracts the multimedia contents according to the candidate group numbers sequentially from the multimedia contents having the highest similarity with reference to a preset candidate group number.
  • the multimedia content is sequentially extracted according to the number of candidates, and four candidate groups of 'if only, romantic holiday, notting hill, and work-to-member' are extracted.
  • the similarity-based multimedia content sorter 550 sorts the multimedia contents extracted according to the candidate group number according to the similarity, and arranges the sorted multimedia contents in the attribute similarity search result output unit ( 600).
  • the Euclidean distance formula is The smaller the similarity value is, the higher the similarity is. Therefore, when the content is rearranged in the order of high similarity, the information is sorted in order of work-to-member, romantic holiday, if only, and notting hill, and the corresponding information is returned to the attribute similarity search result output unit 600. Will be provided to the screen.
  • the multimedia content is obtained when a text keyword search is performed by determining whether to perform a text keyword search or a similar property search by acquiring a search word of the multimedia content input through speech recognition or text. Outputs the search result information of and outputs the search result information of the multimedia contents having the similar property when performing the similar property search.
  • the multimedia content search result most similar to the search word (question) that the user wants to search through is effective.
  • similarity matching analysis when performing a similar attribute search, similarity matching analysis is performed using the attribute information of the search term stored in the search term attribute information DB and the multimedia content attribute information assigned by the multimedia content attribute assignment unit.
  • search result information of multimedia contents having attributes similar to the intention of the search term questions
  • it provides multimedia contents that match the attributes (atmosphere, emotion, etc.) desired by the user, thereby increasing the reliability of the search. Will be effective.
  • Determining whether to perform a text keyword search or a similar property search by acquiring a search word of multimedia content input through speech recognition or text through an apparatus and method for searching multimedia contents through analyzing attribute information according to the present invention. Outputting the search result information of the multimedia content when performing a text keyword search, and outputting the search result information of the multimedia content having a similar property when performing a similar property search, thereby generating multimedia content using a general search keyword method. It is also highly applicable to the industry through providing a multimedia content search result that is most similar to a search word (question) that a user wants to search through the effect of providing a search result and similar property search.

Landscapes

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

Abstract

L'invention concerne un appareil de recherche de contenu multimédia ainsi qu'un procédé de recherche utilisant une analyse d'informations d'attributs et, en particulier, un appareil de recherche de contenu multimédia et un procédé de recherche utilisant une analyse d'informations d'attributs, qui acquièrent un mot de recherche de contenu multimédia entré au moyen d'une reconnaissance vocale ou d'un texte de façon à déterminer s'il faut effectuer une recherche de mot-clé textuel ou une recherche d'attribut similaire, ce qui permet de générer des informations de résultat de recherche de contenu multimédia lorsque la recherche de mot-clé textuel est effectuée, ainsi que de générer des informations de résultat de recherche de contenu multimédia ayant des attributs similaires lorsque la recherche d'attributs similaires est effectuée, et donc de fournir des éléments de contenu multimédia ayant une similarité élevée avec les informations d'attributs de contenu multimédia à rechercher.
PCT/KR2018/002911 2018-03-12 2018-03-13 Appareil de recherche de contenu multimédia et procédé de recherche utilisant une analyse d'informations d'attributs WO2019177182A1 (fr)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
KR1020180028507A KR101873873B1 (ko) 2018-03-12 2018-03-12 속성 정보 분석을 통한 멀티미디어 컨텐츠 검색장치 및 검색방법
KR10-2018-0028507 2018-03-12

Publications (1)

Publication Number Publication Date
WO2019177182A1 true WO2019177182A1 (fr) 2019-09-19

Family

ID=62918154

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/KR2018/002911 WO2019177182A1 (fr) 2018-03-12 2018-03-13 Appareil de recherche de contenu multimédia et procédé de recherche utilisant une analyse d'informations d'attributs

Country Status (2)

Country Link
KR (1) KR101873873B1 (fr)
WO (1) WO2019177182A1 (fr)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111428120A (zh) * 2020-03-17 2020-07-17 北京字节跳动网络技术有限公司 一种信息确定方法、装置、电子设备及存储介质
CN112000822A (zh) * 2020-08-21 2020-11-27 北京达佳互联信息技术有限公司 多媒体资源排序方法、装置、电子设备及存储介质

Families Citing this family (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
KR101913191B1 (ko) * 2018-07-05 2018-10-30 미디어젠(주) 도메인 추출기반의 언어 이해 성능 향상장치및 성능 향상방법
KR20210098135A (ko) 2020-01-31 2021-08-10 주식회사 케이티 질의 데이터를 분석하는 질의 분석 장치, 방법 및 컴퓨터 프로그램
KR102400995B1 (ko) * 2020-05-11 2022-05-24 네이버 주식회사 쇼핑 검색을 위한 상품 속성 추출 방법
KR102399837B1 (ko) * 2020-05-11 2022-05-19 네이버 주식회사 쇼핑 검색을 위한 상품 카테고리 추출 방법
KR102486440B1 (ko) * 2020-11-09 2023-01-09 한국과학기술원 비지도 기반 질의 생성 모델의 학습 방법 및 장치
WO2023074918A1 (fr) * 2021-10-25 2023-05-04 엘지전자 주식회사 Dispositif d'affichage

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
KR20000030847A (ko) * 2000-03-21 2000-06-05 전대식 인터넷 통합서비스 시스템 및 이 시스템에 접근하는 것을용이하게 하기 위한 사용자 인터페이스장치
KR20010028772A (ko) * 1999-09-22 2001-04-06 구자홍 멀티미디어 사용자 프로파일과 사용자 프로파일을 이용한 멀티미디어 검색 및 브라우징 방법
KR20090066608A (ko) * 2007-12-20 2009-06-24 주식회사 다음커뮤니케이션 멀티미디어 컨텐츠 검색 방법 및 시스템
KR100968858B1 (ko) * 2002-04-26 2010-07-09 한국전자통신연구원 사용자 검색 선호도 정보를 이용한 멀티미디어 컨텐츠의 내용 기반 검색 방법 및 장치
KR20100081871A (ko) * 2009-01-07 2010-07-15 포항공과대학교 산학협력단 사용자의 문맥을 바탕으로 개인화된 순위화 검색 방법

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
KR20010028772A (ko) * 1999-09-22 2001-04-06 구자홍 멀티미디어 사용자 프로파일과 사용자 프로파일을 이용한 멀티미디어 검색 및 브라우징 방법
KR20000030847A (ko) * 2000-03-21 2000-06-05 전대식 인터넷 통합서비스 시스템 및 이 시스템에 접근하는 것을용이하게 하기 위한 사용자 인터페이스장치
KR100968858B1 (ko) * 2002-04-26 2010-07-09 한국전자통신연구원 사용자 검색 선호도 정보를 이용한 멀티미디어 컨텐츠의 내용 기반 검색 방법 및 장치
KR20090066608A (ko) * 2007-12-20 2009-06-24 주식회사 다음커뮤니케이션 멀티미디어 컨텐츠 검색 방법 및 시스템
KR20100081871A (ko) * 2009-01-07 2010-07-15 포항공과대학교 산학협력단 사용자의 문맥을 바탕으로 개인화된 순위화 검색 방법

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111428120A (zh) * 2020-03-17 2020-07-17 北京字节跳动网络技术有限公司 一种信息确定方法、装置、电子设备及存储介质
CN111428120B (zh) * 2020-03-17 2023-06-20 北京字节跳动网络技术有限公司 一种信息确定方法、装置、电子设备及存储介质
CN112000822A (zh) * 2020-08-21 2020-11-27 北京达佳互联信息技术有限公司 多媒体资源排序方法、装置、电子设备及存储介质
CN112000822B (zh) * 2020-08-21 2024-05-14 北京达佳互联信息技术有限公司 多媒体资源排序方法、装置、电子设备及存储介质

Also Published As

Publication number Publication date
KR101873873B1 (ko) 2018-07-03

Similar Documents

Publication Publication Date Title
WO2019177182A1 (fr) Appareil de recherche de contenu multimédia et procédé de recherche utilisant une analyse d'informations d'attributs
WO2020009297A1 (fr) Appareil et procédé d'amélioration des performances de compréhension d'un langage sur la base d'une extraction de domaine
WO2018034426A1 (fr) Procédé de correction automatique d'erreurs dans un corpus balisé à l'aide de règles pdr de noyau
WO2010068068A2 (fr) Procédé de recherche d'informations et procédé de fourniture d'informations fondés sur les intentions de l'utilisateur
WO2018174603A1 (fr) Procédé et dispositif d'affichage d'explication de numéro de référence dans une image de dessin de brevet à l'aide d'apprentissage automatique à base de technologie d'intelligence artificielle
WO2012134180A2 (fr) Procédé de classification des émotions pour analyser des émotions inhérentes dans une phrase et procédé de classement des émotions pour des phrases multiples à l'aide des informations de contexte
WO2012074338A2 (fr) Procédé de traitement de langage naturel et de formule mathématique et dispositif associé
WO2020017849A1 (fr) Dispositif électronique et procédé de fourniture de services d'intelligence artificielle sur la base de conversations pré-recueillies
WO2017156893A1 (fr) Procédé de commande vocale et téléviseur intelligent
WO2010021527A2 (fr) Système et procédé d'indexation d'objet dans une image
WO2010036012A2 (fr) Système de recherche d'opinion fondé sur internet, recherche d'opinion, système et procédé de service publicitaire associé
WO2017209564A1 (fr) Procédé de fourniture d'une liste d'applications et dispositif associé
WO2020101108A1 (fr) Plateforme de modèle d'intelligence artificielle et procédé de fonctionnement de plateforme de modèle d'intelligence artificielle
WO2013170662A1 (fr) Procédé et dispositif d'ajout d'informations d'amis, et support de stockage informatique
WO2023172025A1 (fr) Procédé de prédiction d'informations relatives à une association entre une paire d'entités à l'aide d'un modèle de codage d'informations de série chronologique, et système de prédiction généré à l'aide de celui-ci
WO2014021567A1 (fr) Procédé pour la fourniture d'un service de messagerie, et dispositif et système correspondants
WO2011155736A2 (fr) Procédé de production dynamique de termes supplémentaires pour chaque sens de chaque expression en langage naturel ; gestionnaire de dictionnaire, dispositif de production de documents, annotateur de termes, système de recherche et dispositif de construction d'un système d'informations sur des documents basé sur le procédé
WO2020197257A1 (fr) Procédé de traduction utilisant des éléments représentés visuellement, et dispositif associé
WO2012130145A1 (fr) Procédé et dispositif d'acquisition et de recherche d'informations de connaissance pertinentes
WO2020032564A1 (fr) Dispositif électronique et procédé permettant de fournir un ou plusieurs articles en réponse à la voix d'un utilisateur
WO2020022819A1 (fr) Communication par le biais d'un utilisateur simulé
WO2015178716A1 (fr) Procédé et dispositif de recherche
WO2023229376A1 (fr) Système et procédé de recommandation de réponse intelligente pour une assistance de consultation vocale en temps réel
WO2017094967A1 (fr) Schéma de traitement de langage naturel et procédé et système pour établir une base de données de connaissances pour ce dernier
WO2011068315A4 (fr) Appareil permettant de sélectionner une base de données optimale en utilisant une technique de reconnaissance de force conceptuelle maximale et procédé associé

Legal Events

Date Code Title Description
NENP Non-entry into the national phase

Ref country code: DE

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205A DATED 20.04.21)

122 Ep: pct application non-entry in european phase

Ref document number: 18909726

Country of ref document: EP

Kind code of ref document: A1