FR3051936A1 - METHOD AND DEVICE FOR CLASSIFYING MULTIMEDIA CONTENT, TERMINAL AND CORRESPONDING COMPUTER PROGRAM - Google Patents

Abstract

The invention relates to a method for classifying a multimedia content (210, 220, 230) implementing a content classification taxonomy. Such a method comprises the following steps: - customizing the classification taxonomy comprising a selection by the user (20) of at least two categories of interest (21, 22, 23) among the classification categories of the taxonomy; - classification (19) of the multimedia content in the personalized taxonomy (18) comprising: ○ a reproduction to the user, in association, of at least one category in which to classify said multimedia content, and a membership score of the multimedia content to the associated category; ○ a ranking of the multimedia content.

Description

Method and device for classifying multimedia contents, terminal and corresponding computer program. 1. Field of the invention

The field of the invention is that of multimedia navigation, and more particularly that of the classification of multimedia content. Such content can be text (invoices, scientific articles, ...), sound (or audio), images, videos, applications / services or any combination of these elements. The invention applies in particular to the ranking and archiving of multimedia content, whether content generated by a user, or obtained by him. Such a classification can be materialized by a physical storage of multimedia content on a user terminal, or on a remote server, or by a simple annotation of the classified multimedia contents, for example by adding metadata or labels to the contents. The invention may in particular be implemented in a terminal with a user interface and a graphical interface, for example a tablet, a mobile phone, a smartphone ("smart phone"), a personal computer, a connected television to a communication network, etc. 2. Prior art and its disadvantages

Today, when a user wants to store multimedia content (for example an invoice, a photo, a scientific article, a video ...) in a data organization computer system (computer, server, smartphone, or even on the "cloud" ...), it must first identify the most appropriate file for such a ranking, which can be painful and tedious. If, after a search that can be long, it fails to identify appropriate folder, it must create a new folder, whose theme will match that of the content to be classified.

This problem arises, as well for the multimedia contents generated by the user (reports, vacation photos, letters ...), as for the data "consumed" by the user (received invoices, downloaded articles, video to the request...).

This manual classification of content by the user must be document by document, and is a tedious task. In addition, when the user searches for content previously classified, it is common that he has difficulty finding the file in which he had stored. Indeed, it is common that the user forgets the strategy he had adopted for its classification (classification by source, classification by domain ...), and so it is lost in the search for its contents.

Finally, the interests and needs of the user changing over time, It is common that he must change the classification tree created originally, to add new directories, or delete d others; such a fixed classification system therefore lacks flexibility and scalability.

To improve this situation, US Patent 6,621,930 B1 proposes an automatic document classification system based on their textual content. Such a system is based on a preliminary learning phase, making it possible to define, on the basis of a set of learning documents, a set of categories in which to classify the textual documents. When the user wishes to classify a new document, its textual content is analyzed to automatically determine the category, constructed during the learning phase, in which it must be stored. If no appropriate category is detected. It is automatically classified in a neutral category, intended to receive all documents that do not correspond to any of the categories developed during the learning phase.

This technique has several disadvantages. First, it is based on a learning phase, which is long and tedious. In addition, the classification tree proposed to the user is frozen at the end of this learning phase, and depends greatly on the diversity and the number of learning documents used: if the user's documents are categories that were not created during the learning phase (because none of the learning documents related to them), they will all be automatically ranked in the neutral category, which not relevant to the user. This technique therefore does not allow the user a personalized classification of his documents, which is adapted to his needs.

US Patent 7,047,236 B2 also provides a technique for automatically inferring rules for matching content and rating categories. This rule setting is based on finding keywords in content, and ranking content into categories, based on the keywords they contain, and an occurrence score. Again, this technique is based on a ranking tree that is fixed and not scalable. It does not allow the user to customize the classification into categories of its contents, and can therefore be tedious to use if the proposed tree is not adapted to the interests of the user.

There is therefore a need for a technique for classifying and classifying multimedia contents that does not have these various disadvantages of the prior art. In particular, there is a need for such a technique, which allows the user a customization of its classification, and which is scalable. 3. DISCLOSURE OF THE INVENTION The invention responds to this need by proposing a method of classifying a multimedia content implementing a content classification taxonomy. Such a method comprises the following steps implemented by a processor: a prior step of customizing the classification taxonomy comprising: O a presentation to a user of a plurality of classification categories of the taxonomy; O a selection by the user of at least two categories of interest from the plurality of categories presented; a step of classifying the multimedia content in the custom classification taxonomy comprising: O a rendering to the user, in association with: at least one category of the custom classification taxonomy in which to classify the multimedia content, and a score the membership of the multimedia content to the associated category; O a ranking of the multimedia content in a category selected from: the category or categories returned to the user; another category of the custom classification taxonomy chosen by the user.

Thus, the invention is based on a new and inventive approach to the classification of multimedia content. Indeed, the invention proposes to the user to be able to customize the classification taxonomy of its multimedia content, so that it is perfectly adapted to its needs and interests, while offering an automatic classification tool allowing him to easily determine which category, among those he has selected, is the most suitable for the content to be classified.

Thus, the user can for example retain categories "Arts", "Society", "Sciences" and "Space and Time", and exclude categories "Technologies" and "Person and Character", which do not correspond to its centers interest.

In addition, the technique of the invention automatically restores to the user the category or categories that best correspond to the multimedia content to be classified, with a membership score of the content to the proposed category. Thus, for example, the user sees on his screen that the content he wants to classify corresponds to 60% to the category "Tourism", 30% to the category "sport", and 10% to the category " food ". The number of categories returned to the user with their membership score can be set from 1 to N, or chosen according to the targeted application. The user can then follow the recommendation of classification, by classifying its content in the category presenting the highest membership score, or choose one of the other categories proposed, or opt for a category that has not been proposed, but which seems to him preferable. The user therefore retains flexibility in the final choice of the category in which to classify its multimedia content, while enjoying a significant automatic classification help. When the user is then looking for previously ranked content, he can more easily identify the category in which he was ranked, if he has made the choice to classify it in the category with the highest membership score. Student.

By "category" is meant a logical division of the classification taxonomy.

By "taxonomy" is meant an organizational scheme of classifying elements, in this case multimedia content, in the form of a sequence of elements forming lists. In the present case, the classification taxonomy within the meaning of the present invention comprises a hierarchy of categories, in which each lower level of the hierarchy comprises subsets ("daughter subcategories") of the categories of the directly higher level.

According to one embodiment, the personalization step of the classification taxonomy comprises for each of the selected categories of interest, a choice by the user of a depth level of the classification taxonomy making it possible to determine a number of levels. hierarchical sub-categories-girls for the category of interest.

Thus, such customization allows the user, on the one hand, to choose a set of "mother" categories, among the existing categories of the classification taxonomy, and on the other hand, for each of these chosen mother categories, define the number of hierarchical levels of sub-categories-girls that are needed.

In a category "tourism", which he practices a lot, he can foresee a deep tree of subcategories-girls, to allow a precise classification of all his travel documents (for example, a sub-category-girl by continent "tourism in Europe", "tourism in Asia", "tourism in Africa", "tourism in North America", "tourism in Oceania", "tourism in South America", then in each of these subcategories girls, new sub-categories-girls by country visited, etc.). Conversely, in the "cinema" category, which he is not keen on, he will be able to decide not to foresee a sub-category-girl, and to arrange all his contents in the general category "cinema". For purposes of lightening terminology, the term "category" is used in the remainder of this document to refer to parent categories, located at the root of the classification taxonomy, but also to designate subcategories. girls, regardless of their depth level in the taxonomy tree.

According to one embodiment of the invention, the classification step also comprises: a determination of at least one key-data present in the multimedia content; an identification of at least one category and / or sub-category-daughter of the personalized taxonomy with which the data (s) -clef (s) are associated with a corresponding relevance score; a calculation, for each identified category and / or daughter subcategory, of the membership score of the multimedia content in the category and / or sub-category-girl, from the relevance score of the data or data associated with the category and / or sub-category-girl.

By "key-data" we mean relevant information contained in the multimedia content, or deductible from the latter, and allowing to link the content to a category of the taxonomy. For multimedia content that is a textual document, such key-data may be a keyword; for a photo, it may be the name of the monument it represents, for example.

Thus, the key data defining the multimedia content to be classified are extracted from the content. They are compared to the weighted key data that define the different categories of the taxonomy. By summing the relevance scores obtained for each key-data item for a given category, the membership score of the multimedia content is determined for this category.

The automatic classification proposed to the user is therefore based on reliable and reproducible statistical data.

According to one embodiment of the invention, the content classification taxonomy is constructed from a scalable collaborative database. For example, the content classification taxonomy is derived from the DBpedia® collaborative database, which is based on the collaborative Wikipedia® platform. Thus, this approach provides a classification taxonomy that is based on the most comprehensive list of user categories and interests, and covers all areas, however diverse. Each taxonomy category can then be defined by a set of weighted keywords extracted from Wikipedia® articles, such as the anchors of these articles.

In addition to the comprehensiveness of the areas covered, this approach allows the construction of a dynamic and evolutionary classification taxonomy, both in terms of the number of categories of taxonomy (automatic addition of new categories, according to new published articles), and in the ranking of a content in a category (through the appearance of new keywords), or that the calculation of the relevance score of a keyword for a category (which may change over time, depending on the appearance of new articles or new keywords).

Such an evolutionary taxonomy can therefore make it possible to constantly change the classification of the user's content, and even to adapt the classification of content previously archived according to the evolution of the taxonomy.

According to one embodiment, such a method also comprises a step of determining at least two points of interest of the user among the categories and / or sub-categories-girls of the custom classification taxonomy, according to scores. belonging to N multimedia contents, where N is an integer greater than or equal to 2, to the categories and / or subcategories-girls, and a step of presenting to the user of the centers of interest determined, in the form of a list of categories and / or sub-categories-girls associated with an affinity factor calculated from the membership scores.

Thus, depending on the categories in which the user's multimedia content N has been classified, it is possible to deduce what are the user's interests and to give him this information. The higher N is, the greater the reliability of the identification of the centers of interest.

According to a first embodiment, the affinity factor Q of one of the points of interest C is calculated according to the equation:

where N is the number of multimedia contents classified in the category and / or subcategory girl C, and Sc, i is the membership score of the i-th multimedia content to the category and / or subcategory daughter C.

According to a second embodiment, the affinity factor Q of one of the points of interest C is calculated according to the equation:

where N is the number of multimedia contents classified in the category and / or subcategory girl C, Se, i is the membership score of the i-th multimedia content to the category and / or subcategory girl C and dj is a number of days since the classification of the i-th multimedia content.

Thus, the time factor can be integrated in the user's interest assessment, so as to give greater importance to the multimedia content that he has recently classified.

According to one embodiment, such a method also comprises a step of forming a group of users having common centers of interest and associated associated affinity factors.

Indeed, based on the points of interest determined for a plurality of users, and associated affinity factors, it is possible to compare the users, and to create groups or "clusters" of users presenting profiles. Similar. For example, we can create a group of users all having as centers of interest the categories "museums", "sculpture" and "painting" with an affinity factor greater than or equal to 15%.

It is thus possible to offer users to be part of a community of users, for example through social networks. This constitution can also be used in groups of users to target promotions or commercial offers to a particular group of users. It can also be used to customize services and offers based on user profiles.

The various embodiments or aforementioned embodiments can be added independently or in combination with each other, to the steps of the method of ranking a multimedia content defined above. The invention also relates to a device for classifying a multimedia content implementing a content classification taxonomy. Such a device comprises: a personalization module of the classification taxonomy comprising: O a presentation interface to a user of a plurality of classification categories of the taxonomy; O a user selection interface of at least two categories of interest among the plurality of categories presented; a multimedia content classification module in the custom classification taxonomy comprising: O an interface for rendering the user, in association: at least one category of the custom classification taxonomy in which to classify the multimedia content, and a score of membership of the multimedia content in the associated category; O a categorization module of the multimedia content in a category selected from: the category or categories returned to the user; another category of the custom classification taxonomy chosen by the user.

According to one embodiment, the selection interface of the personalization module of the classification taxonomy is also configured to allow the user, for each of the selected categories of interest, the choice of a depth level of the taxonomy. classifying a number of hierarchical levels of sub-categories-girls for the category of interest. The invention also relates to a terminal comprising the multimedia content ranking device mentioned above.

Such a terminal is for example a computer, a tablet or a smartphone, possibly connected to a content storage server. The invention also relates to a computer program comprising program code instructions for performing the steps of the method of classifying multimedia content according to the invention, when executed by a computer.

Such a program can use any programming language, and be in the form of source code, object code, or intermediate code between source code and object code, such as in a partially compiled form, or in any another desirable form. The invention also relates to a computer-readable recording medium on which is recorded a computer program comprising program code instructions for performing the steps of the method of classifying a multimedia content according to the invention such that as described above.

Such a recording medium may be any entity or device capable of storing the program. For example, the medium may include storage means, such as a ROM, for example a CD ROM or a microelectronic circuit ROM, or a magnetic recording means, for example a USB key or a hard disk. On the other hand, such a recording medium may be a transmissible medium such as an electrical or optical signal, which may be conveyed via an electrical or optical cable, by radio or by other means, so that the program computer that it contains is executable remotely. The program according to the invention can in particular be downloaded to a network, for example the Internet network.

Alternatively, the recording medium may be an integrated circuit in which the program is incorporated, the circuit being adapted to execute or to be used in the execution of the aforementioned display control method.

The device for classifying multimedia content, the aforementioned terminal and the corresponding computer program have at least the same advantages as those conferred by the process of classifying a multimedia content according to the present invention. 4. List of Figures Other purposes, features and advantages of the invention will emerge more clearly on reading the following description, given as a simple illustrative and non-limiting example, in relation to the figures, among which: FIG. 1 illustrates the method of constructing the knowledge base defining categories of the classification taxonomy according to one embodiment of the invention, as well as the use of this knowledge base to classify a multimedia content of the user; FIG. 2 provides a block diagram of the general principle of the method of classifying multimedia contents according to one embodiment of the invention; FIG. 3 describes in a functional flowchart the various steps implemented by the method of FIG. 2; FIG. 4 shows the simplified structure of a device for classifying multimedia contents, according to an exemplary embodiment of the invention. 5. Detailed Description of Embodiments of the Invention

The general principle of the invention is based on the implementation of a taxonomy of content classification scalable and customizable by the user, to offer him a tool for automatic classification of its multimedia content, perfectly suited to his needs and his interests.

An embodiment of the construction of the knowledge base defining the categories of the classification taxonomy according to the invention, based on the use of the DBpedia® database, is now presented with reference to FIG. which defines the structure of the collaborative platform Wikipedia®. Each category (or sub-category-girl) of the classification taxonomy is defined by a set of weighted keywords extracted from Wikipedia® articles. These keywords can be concepts, or named entities, and the score associated with them represents their degree of interest in each category.

Some keywords can be repeated in several categories, but associated with different scores, which reflect a different degree of interest for each of these categories.

The Wikipedia® collaborative platform includes the following parent categories:

Arts;

Science;

Space and Time;

Society ;

technology;

Person and Character.

As illustrated in FIG. 1, a first COLLEC step. referenced 10 allows to collect the data necessary for the definition of the knowledge base. First, we extract from the DBpedia® database the categories and subcategories-girls it lists, for example using a SPARQL query. It is recalled that SPARQL is a query language and a protocol that makes it possible to search for, add, modify or delete RDF data (available for the Resource Description Framework) available through the Internet. Its name is a recursive acronym that stands for SPARQL Protocol and RDF Query Language.

For example, the following SPARQL query retrieves all -1 level child subcategories of the parent category "Company": select? Cat? CatLabel? ImmediateParent? Level where {? Cat a skosiConcept. ? cat rdfsilabel? catLabel. {{? cat skosibroader <http://en.dbpedia.org/resource/Category:Company>. ? cat skosibroader? immediateParent. BIND (1 AS? Level) order by? Level? ImmediateParent? Cat

After collection of all the categories and subcategories-girls, it then collects during the referenced step 10 COLLECT WIKIPEDIA® articles that define them, again by means of a query SPARQL. For example, the following SPARL query retrieves Wikipedia® articles for the "Company" category: select? Uri? WikiPage where {? Uri dctermsisubject <http://en.dbpedia.org/resource/Category : Society>. ? uri foafiisPrimaryTopicOf? wikiPage. }

Finally, again during the COLLECT 10 step, we extract from all the collected Wikipedia® articles all the anchors they contain. It is recalled that anchors are the hyperlinks contained in Wikipedia® articles, and are keywords that may represent concepts, or named entities. Thus, according to one embodiment of the invention, we do not use the entire content of Wikipedia articles, but only the anchors they contain, also called tags.

Thus, by applying this method to the parent category "Company", and by collecting all the sub-categories-girls up to a level of tree depth of -4, COLLECT 10 is collected. , 49929 separate subcategories, 804,722 Wikipedia® articles and about 45 million anchors, or tags. As will be seen later, these anchors are used to define the categories and subcategories of the classification taxonomy. The set of information collected during the COLLECT step 10 is stored in a database referenced 11. At the end of this information collection step, a pre-filtering step FILTR 12 makes it possible to delete, iis tagged tags, keywords that are meaningless for the definition of different categories, ie numbers, dates, country_lists ...

The anchors obtained after filtering FILTR 12 form a set referenced 13 in FIG. 1. This filtering 12 makes it possible to save in processing time during the subsequent step of SCORING 14, which is therefore applied only to the set 13 keywords with meaning.

During this step of SCORING 14, a variant of the TF-IDF algorithm is used to calculate the relevance score of each keyword for a given category.

It is recalled that the TF-IDF (Term Frequency-Inverse Document Frequency) algorithm is a weighting method, which provides a statistical measure to evaluate the importance of a term contained in a document, in relation to a body of documents. The weight, or score, assigned to the term considered, increases proportionally to the number of occurrences of the word in the document. It also varies according to the frequency of the term in the corpus.

More precisely, during the step SCORING 14, the following score is calculated for a term keyword:

NOT

Score (term) = tfx log- lii Where: tf is the number of occurrences of each word in articles

Wikipedia® belonging to the category and associated sub-categories-girls, N represents the total number of Wikipedia® articles for these categories Π; is the number of occurrences of the keyword term in the list of articles

Wikipedia®.

After calculation of the relevance scores of all the anchors of the set 13, a normalization is performed on all the scores of relevance obtained. The purpose of this standardization is twofold. On the one hand, it provides a standardized relevance score that reflects the degree of interest of a keyword, or term, relative to other keywords in the same category.

For example, the standardized relevance score defines the importance of a term in a particular category as a percentage. For example, if the term "hotel" gets a standardized score of 0.01 in the "Tourism" category, this means that the term "hotel" represents 1% of the sum of the relevancy scores of all the terms for this category " Tourism ". On the other hand, such a standardization makes it possible to obtain a common reference base for all the categories of the classification taxonomy, which makes it possible to evaluate the relative importance of a term in the different categories in which it appears. . Thus, if the term "hotel" belongs to both the "tourism" and "food" categories, with respective standardized relevance scores of 0.01 and 0.001, it can be inferred that the keyword "hotel" is more important. in the category "tourism" (where it represents 1% of category relevance scores) than in the "food" category (where it represents only 0.1% of category relevance scores). All of the tags, and their associated normalized relevance scores, are stored in a referenced database 15.

During the OPTIM step referenced 16, one selects, for each of the categories considered, the set of anchors of the database 15 having the highest standardized relevance scores. For example, for each category, only the 100 or 500 anchors with the highest relevance scores for a category, or the anchors whose relevance score is greater than a certain relevance threshold, are retained. This optimization OPTIM 16 allows to keep only a list of anchors most representative of the category, and has a significant impact on the performance of the classification process. At the end of this selection OPTIM 16, the INDEX step referenced 17 makes it possible to lighten the subsequent process of classification of the multimedia contents of the user 20, by adopting a different representation of the categories of the classification taxonomy. Indeed, at this stage, we have a list of categories and, for each category, a list of main keywords weighted by their relevance score. The indexing step INDEX 17 aims on the contrary to obtain a list of keywords, with, for each keyword, all the categories in which it is present, in association with the standardized relevance score of the word- key in each of these categories. The result of the indexing step INDEX 17 can be presented for example in the form of the table referenced 18 in FIG. 1. At this stage, one thus has the knowledge base underlying the document classification taxonomy. implemented in one embodiment of the invention.

It is then possible, during a CLASSIF 19 step, to classify the multimedia contents of a user 20, using this knowledge base.

This classification CLASSIF 19 will now be described in more detail in relation to FIGS. 2 and 3. FIG. 2 provides a block diagram of the general principle of the method of classifying multimedia contents according to one embodiment of the invention. FIG. 3 describes in a flowchart form the various steps implemented by this method.

As illustrated in FIG. 2, the user 20 has a certain number of multimedia contents, among which textual contents 210, photos 220, and videos or films 230. The automatic classification of these contents 19 uses the base knowledge of 18 constructed according to the method of Figure 1, to assign each of these contents to one or more categories of the classification taxonomy, for example: category I: Tourism, referenced 21; category II: Cinema, referenced 22; category N: Sport, referenced 23.

More specifically, when the user wishes to classify a multimedia content, for example a textual content 210, he first begins by customizing the content classification taxonomy elaborated in accordance with FIG. 1 during a CUSTOM step. .

To do this, it displays on the screen of its terminal (for example a tablet or a personal computer), a simple representation of the classification taxonomy tree, during a PRES stage. CAT. For example, at first, only the parent categories of the taxonomy are displayed. The user can then select, during a SELECT 302 step, those categories that he wishes to keep in his classification tree. This selection can be made for example by means of checkboxes. For each mother-category selected, then on the screen of the user 20, the lower-hierarchically hierarchical daughter subcategories are displayed, from which the user can once again choose which ones he wishes to keep or not. This is done until the user has fully customized the classification taxonomy, having determined on the one hand the parent categories he needed, and on the other hand, the level of hierarchical depth of sub- girls categories for each of these parent categories. At the end of this step SELECT 302, a set of Cat categories is thus obtained. Cat. N, with depth level Prof. 1 to Prof. N associated.

This customization step of customizing the classification taxonomy can be done once, upon initialization of the content classification method according to one embodiment of the invention.

As a variant, it can also be proposed to the user 20 with each new multimedia content that he wishes to classify. Indeed, the classification taxonomy of content based on a collaborative platform scalable Wikipedia® type, it can evolve dynamically, and offer new categories from time to time, as new Wikipedia articles and new associated keywords. It may therefore be interesting for the user to always check that the classification tree he had previously selected is still appropriate for him, and that he does not wish to add new sub-categories-girls or delete others become obsolete.

It can also be envisaged that this customization step of the classification taxonomy is proposed to the user from time to time, for example at regular expiry (eg every month) or at the end of the classification of a number of determined multimedia content (eg all 10 contents).

When the user 20 creates or downloads on his terminal a multimedia content 211, for example an invoice received from a hotel in which he has stayed, this content 211 (in this example, textual content) is submitted to the CLASSIF step. 19 automatic classification of content.

During this step CLASSIF 19, the key data present in the content 211 is first identified. These key data can be keywords (for example: bill, hotel, room, breakfast, etc.). ), or the hotel logo. For a photo-type multimedia content, the key-data may be the identity of persons present in the photo, identified by means of face recognition software associated with a bank of faces, or the name of a monument appearing in the photo, or a keyword identified by character recognition on the photo (for example a city name on a traffic sign). For video content, the key data can be extracted from metadata associated with the content (for example, the name of the film, the name of the actors, the name of the director ...).

In addition, when the multimedia content 211 is a textual document, it is possible to choose to use all the words that appear in the document or, alternatively, to apply a word weighting algorithm, for example of the TF-IDF type, to retain only a few important keywords.

These key data identified in the content 211 can also be called tags.

During a step referenced PAIR 191, a matching function is applied between the keywords extracted from the content 211 during the DETER step. TAGS 190 and tags tagl, tag2, ... of the knowledge base 18, to identify the category or subcategories cat. 1 to cat. N of the custom taxonomy associated with these keywords, with their associated relevancy score.

During a step CALC SCORE 192, for each of these categories, a membership score of the content 211 is calculated for this category, on the basis of the relevance scores of the different keywords of the content 211. For example, to determine the membership score of the content 211 to the category cat. 1, we sum the relevance scores of all the keywords of the content 211 associated with the cat category. 1.

During a RESTIT step. 193, the user is then returned the result of the steps referenced 190 to 192. To do this, several variants can be envisaged: it can be presented to the user 20, for example by displaying on the screen of his terminal the category closest to the multimedia content 211, ie the one which obtained the highest membership score after the step 192; it is possible to present to the user 20 all the categories identified during the step 191, in association with their membership score calculated during the CALC stage SCORE 192, for example ranked in descending order of the membership score ; the user can be presented with a subset of the identified categories, for example the three categories with the highest membership scores.

For example, during steps 190 to 192, it is determined that the multimedia content 211 relates to: the category "Tourism" with a membership score of 60%; the category "Sport" with a membership score of 30%; the category "Food" with a membership score of 10%.

This result is displayed on the screen of the terminal of the user 20. The latter can then choose in which category he wishes to classify the content 211 during a CLASS step referenced 194. To do this, he can follow the recommendation automatic classification, ranking the content in the category "Tourism" which has obtained the highest membership score. He may also choose the category "Sport", which has a lower membership score, but seems preferable. He may also be content to take note of this recommendation, but opt for another category of the custom classification taxonomy, which seems to him to better meet his expectations or classification logic.

Such a classification can be associated with electronic archiving of the content in a dedicated directory of its terminal, or a remote storage server. It may also consist of adding metadata to the content, indicating the classification category chosen.

The user is thus offered valuable assistance for the automatic classification of his multimedia contents, which becomes simple and logical. The search for previously classified content is also facilitated.

Indeed, the classification taxonomy used according to one embodiment of the invention relies on a collaborative platform to define a very diverse list of categories, which are not restricted to a particular field, but instead cover all the cultural domains, technical, historical, everyday ... In addition, the approach of the classification method according to one embodiment of the invention is dynamic, because it relies on a collaborative platform, which is permanently powered and put up to date, either by new subcategories or by new keywords associated with these subcategories. The degree of importance of each keyword in a category, or relevance score, is also dynamic and evolutive, since the result of its calculation may vary depending on the appearance of the keyword considered in new articles associated with the category.

In relation to FIG. 4, we now consider the simplified structure of a device 100 for classifying multimedia contents, according to an exemplary embodiment of the invention. Such a multimedia content ranking device is adapted to implement the multimedia content classification method previously described in relation with FIGS. 2 and 3.

In this example, the device 100 is integrated in a tablet, a computer or any other TER user terminal.

It comprises a processing unit 110, equipped for example with a processor P, and driven by a computer program PG 120, stored in a memory M 130 and implementing the method of classifying multimedia contents according to an embodiment of the invention. The TER tablet includes a communication unit UC which contains an emisslon / reception module adapted to transmit and receive the information exchanged with the personalization module PERS of the classification taxonomy by the user. At initialization, the code instructions of the computer program PG 120 are for example loaded into a RAM (not shown) before being executed by the processor P of the processing unit 110. The processor of the processing unit 110 implements the steps of the multimedia content classification method described above, according to the instructions of the computer program 120.

In a manner known per se, the tablet TER comprises: an INT module for processing user interactions U1, an EC display screen.

According to the invention, the device 100 for classifying multimedia content also comprises a personalization module PERS of the classification taxonomy, comprising a display unit of the presentation interface PRES to the user of the categories of the taxonomy of classification, and SELEC selection interface by the user of the categories and subcategories that he wishes to retain in the custom classification taxonomy. It also includes a CLASSIF module for automatic classification of multimedia content, comprising an RES module of content membership categories, with their membership score, and a CLASS classification module of the content. The modules PERS and CLASSIF are directly controlled by the processor P of the processing unit 110.

The automatic content classification proposed according to one embodiment of the invention can make it possible to identify the interests of a user, i.e. the categories of the custom classification taxonomy that he uses the most. To identify these points of interest, we can use the formula:

where N is the number of multimedia contents classified in category C, and Se_i is the membership score of the i-th multimedia content in category C.

One may also wish to integrate the time factor in the calculation of the centers of interest, by adopting the alternative formula:

where N is the number of multimedia contents classified in category C, Se ^ i is the membership score of the i-th multimedia content in category C and dj is a number of days since the classification of the i-th multimedia content . This makes it possible to give greater importance to content classified more recently.

This calculation makes it possible to obtain the interests of the user, in the form of a list of categories with their affinity factors, for example:

Ci = {18% tourism 12% travel 8% theater 3% sport 1,8% cinema} From the calculation of users' interests, we can establish a matrix of interests for all users using automatic document classification service according to an exemplary embodiment of the invention, for example:

Catl Cat2 catn

Userl 18% 12% 7%

User2 2% 23% 8%

Usern 15% 1.2% 2%

This matrix calculates the affinity (or proximity) between users based on their interests. It is expressed in a list of categories with similar affinity factors. The number of categories to be taken into account can be parameterizable, for example at least three categories. Thus, one can form groups or "clusters" of users with neighboring interests.

For example, a first Group 1 group of users who like museums, sculpture, art and painting, and a second Group 2 users who love tourism, travel, cruises ...

Group 1: {

Users: [user58, userll, user200 ...]

Categories: [museums, sculpture, art, painting ...]}

Group 2: {

Users: [userS, user37, userlll ...]

Categories: [tourism, travel, cruises, hotels ...]}

This calculation of affinity (or proximity) between users can allow, among other things: • To offer users to become part of a community, for example through social networks; • Target promotions to a group of people with common interests; • Customize services and offers

Claims (10)

CLAIMS AMENDED A method of classifying a multimedia content (210, 220, 230, 211) implementing a content classification taxonomy, characterized in that said method comprises the following steps implemented by a processor: a preliminary step of customizing (30) said classification taxonomy comprising: O presenting (301) a user (20) of a plurality of classification categories of said taxonomy; O a selection (302) by said user of at least two categories of interest (21, 22, 23) among said plurality of categories presented; O for each of these selected categories of interest (cat.1, cat.N), a choice by said user of a depth level (prof.1, Prof.N) of said classification taxonomy making it possible to determine a number of hierarchical levels of sub-categories-girls for said category of interest, a step of classifying (19) said multimedia content (211) in said custom classification taxonomy (18) comprising: O a determination (190) of at least one key-data present in said multimedia content (211); O an identification (191) of at least one category and / or sub-category-daughter of said personalized taxonomy with which said at least one key-data item is associated with a corresponding relevance score; O a calculation (192), for each category and / or subcategory-girl identified, of said membership score of said multimedia content to said category and / or sub-category-girl, from the relevance score of said at least one key data associated with said category and / or sub-category-girl; O a reproduction (193) to the user, in association: of at least one category of said custom classification taxonomy in which to classify said multimedia content, and of a membership score of said multimedia content to said associated category; O ranking (194) said multimedia content in a category selected from: said at least one category returned to the user; another category of said custom classification taxonomy chosen by the user. 2. Classification method according to claim 1, characterized in that said content classification taxonomy is constructed from a scalable collaborative database. 3. A method of classification according to claim 1 or 2, characterized in that it also comprises a step of determining at least two centers of interest of said user among said categories of said custom classification taxonomy, according to the scores of d. belonging to N multimedia contents, where N is an integer greater than or equal to 2, to said categories, and a step of presenting to said user of said at least two specified points of interest, in the form of a list of categories associated with a factor affinity calculated from said membership scores. 4. Classification method according to claim 3, characterized in that said affinity factor Cj of one of said centers of interest C is calculated according to the equation: where N is the number of multimedia contents classified in said category C, and Sc, i is the membership score of the i-th multimedia content to said category C. 5. Classification method according to claim 3, characterized in that said affinity factor Q of one of said centers of interest C is calculated according to the equation: where N is the number of multimedia contents classified in said category C, Sc, i is the membership score of the i-th multimedia content to said category C and dj is a number of days elapsed since the classification of the i-th multimedia content . 6. Classification method according to any one of claims 3 to 5, characterized in that it also comprises a step of forming a group of users having common centers of interest and associated affinity factors close . 7. Device for classifying a multimedia content implementing a content classification taxonomy, said device being characterized in that it comprises: a personalization module (PERS) of said classification taxonomy comprising: ο a presentation interface (PRES) to a user of a plurality of classification categories of said taxonomy; O a selection interface (SELEC) by said user of at least two categories of interest among said plurality of categories presented; O an interface of choice by said user of a depth level (Prof. 1, Prof.N) of said classification taxonomy making it possible to determine a number of hierarchical levels of subcategories-girls for said category of interest, for each of said selected categories of interest (cat.1, cat.N), a classification module (CLASSIF) of said multimedia content in said personalized classification taxonomy comprising: O a module for determining at least one key-data item present in said multimedia content (211); O a module for identifying at least one category and / or sub-category-daughter of said personalized taxonomy with which said at least one key-data item is associated with a corresponding relevance score; O a calculation module, for each category and / or subcategory-girl identified, of said membership score of said multimedia content to said category and / or sub-category-girl, from the relevance score of said at least one key data associated with said category and / or sub-category-girl; O an interface (RES) to restore to the user, in association: at least one category of said custom classification taxonomy in which to classify said multimedia content, and a membership score of said multimedia content to said associated category; O a classification module (CLASS) of said multimedia content in a category chosen from: said at least one category returned to the user; another category of said custom classification taxonomy chosen by the user. Terminal (TER) capable of containing a device for classifying a multimedia content implementing a content classification taxonomy according to claim 7. A computer program comprising program code instructions for performing the steps of the media classification method implementing a content classification taxonomy according to any one of claims 1 to 6, when said program is executed by a processor. A computer-readable recording medium on which a computer program is recorded with program code instructions for performing the steps of the method of classifying multimedia content implementing a content classification taxonomy. according to any one of claims 1 to 6, when said program is executed by a processor.