FR2971597A1 - Method for calculation of social relevance index of digital documents to sort list of documents to search documents in documentary base in enterprise, involves calculating index from set of actions realized on document - Google Patents

Abstract

The method involves calculating (21) a proximity value of a first set of users (U1) from a set of data included in a user profile. A second set of users (U2) whose proximity value is higher than a predetermined threshold, is selected (22) from the first set of users, where the calculated proximity value is dependant on the first set of users. Social relevance index of documents is calculated from a set of actions realized on the documents by the second set of users using a calculator. Independent claims are also included for the following: (1) a method for sorting a list of documents (2) a method for searching documents in a documentary base (3) a system for calculation of social relevance index of a digital document (4) a computer program comprising a set of instructions for implementing a method for calculation of social relevance index of a digital document.

Description

METHOD OF CALCULATING A PERTINENCE INDEX OF A DOCUMENT FROM A DATABASE, METHODS OF SORTING AND SEARCHING THE SAME, SYSTEM FOR IMPLEMENTING METHODS, AND ASSOCIATED COMPUTER PROGRAM. The present invention relates to methods of searching a document base. More particularly, the invention relates to a method of calculating a social relevance index of a document of a document base for a given user. Finally, the field of the invention applies to a closed network of users. Currently, many sorting methods make it possible to return a list of relevant documents according to predefined search criteria. Most search engines, such as Google, for example, render a list of corresponding links to websites, sorted by a relevance criterion, from a query defining keywords. The sorting performed depends on a search mode taking into account in particular the number of occurrences of a keyword in the content of the website. Other parameters are used but mainly involve the correlation between the keywords defined in the request and the content of pages of websites. In addition, links between the pages are also used to determine the pages appreciated by other sites. Most of the sorting and document retrieval methods rely on the creation of an index grouping key words and themes that are referenced in the documents of a database. Thus, a document search method from the definition of a query does not scan the contents of all the documents and may include a search step in the index to find and sort the relevant documents. FIG. 1 represents such a method making it possible, from predefined subjects, to generate automatic requests from a recommendation system during a user's navigation. The search engine then searches for an index comprising themes or keywords or keywords collected from a database of documents Dn. The interaction between the request and the index being defined by means of a matching step A. Other sorting methods exist and include a marking of the documents made by users. These methods make it possible to take into account a dimension of appreciation in the sorting and especially if a user has appreciated a document. These methods are notably used in the search for media of certain websites which make it possible to highlight or sort documents according to a criterion of appreciation of a medium.

Other document sorting methods include a definition of user membership groups and the assignment of certain rights to these groups. These rights include different types of possible actions on documents. Some document management systems use sorting methods that integrate the interactions between documents and users in the assignment of relevance index to the documents returned from a request, this is the case of the system described in patents US7761471 and US20020087600. A disadvantage of current sorting methods is that their performance is expensive in terms of calculations. In addition, the resources required for the calculations become very important when a document sorting process takes into account social data, in particular the data of each user profile and the interactions between these profiles and the semantic taggages of the documents. As a result, a drawback of the sorting methods is that they do not take into account the correlations that may exist between the users or minimize them so as not to penalize the computation resources. Another disadvantage is that some data between a document and a user, such as data representing appreciations, is stored in a database but only partially reflects the common interests of the different users. Indeed, the user profiles can include different types of data that are not taken into account in the sorting criteria to define a relevance index of a document.

A problem that arises is therefore to improve the performance of digital document sorting processes. The present invention proposes to improve document sorting processes in order to optimize document searches in the context of using a search engine. The invention solves these disadvantages. In addition, the invention makes it possible to define a subset of users close to a first user performing a search from the calculation of a proximity value with each user.

The invention makes it possible to take into account the proximity value with the calculation of a relevance index of a document from a request defined by the first user, the correlation of the proximity value with the relevance index. defining an index of interest. The set of documents are sorted by the method of the invention by a scheduling according to each index of interest. Advantageously, the method of calculating a social relevance index of a first digital document accessible to a first set of users, each user being able to perform a first set of actions on the first document, comprises: a step identifying a first user of the first set; a step of calculating at least one proximity value with the first user from a set of data included in a user profile of each user; a step of selecting in the first set of a second set of users whose proximity value is greater than a predetermined threshold, each calculated proximity value being dependent on the first user; a step of calculating, from the calculator, the social relevance index from the set of actions performed by each user of the second set on the first document. Advantageously, the first set of actions includes the modification of a document, the creation of a document and the consultation of a document. Advantageously, the first set of actions comprises generating a measurable judgment of a document. Advantageously, the first set of actions comprises the transmission of a document by means of an electronic mail. Advantageously, the first set of actions includes a download of a document or a selection of a part of the document. Advantageously, the invention relates to a method for sorting a list of documents, comprising: a step of calculating, from a calculator, a set of indices of social relevance according to the method, of a first list of documents ordered according to a first order; a step of sorting, from a calculator, the first list of documents ordered according to a second order corresponding to a descending order of the value of each index of social relevance. Advantageously, each document comprising a semantic descriptor referencing a set of semantic data specific to each document comprising: a step of defining a first request defined by means of a user interface, the user defining the request making it possible to determine the first user of the method, said request comprising at least one term; a step of generating, from the computer, a second list of digital documents extracted from a first documentary database, the generation of the second list being triggered from the first request, the second list being ordered from a first function of generating a semantic relevance index attributing to each document a value resulting from a correlation between the terms of the request and the semantic descriptor of each document, the second list being ordered according to a descending order of the value of the semantic relevance index, the second list including the first list of documents.

Advantageously, a user profile of a user includes a list of the actions performed on each of the documents of the first document base. Advantageously, a user profile of a user comprises a set of transmission of the documents of the first document base and the recipients associated with the transmissions. Advantageously, a user profile comprises an interest profile comprising semantic data including a list of terms. Advantageously, the proximity value of a user profile of a user of the second set with the first user is calculated from a second function determining a first value calculated from a correlation between the terms of the interest profile. of each user of the first set with the interest profile of the first user. Advantageously, the list of terms referenced in each interest profile comprises the number of occurrences of each term referenced several times in the interest profile of each user of the first set, the second function weighting the common terms between the profile of each user. interests of each user of the first set with the interest profile of the first user by their number of occurrences, a second value being calculated. Advantageously, the interest profile of a user of the first set is generated from a third function making it possible to extract a list of terms from the semantic descriptor of at least one document on which the user has performed an action of the first set of actions and weight each extraction by the action type on the document. Advantageously, the user profile comprises a fourth value corresponding to the result of a fourth function comparing a status of a user with the first user, each user comprising a status, the different statuses being differentiated by a value representing a degree of difference. Advantageously, the proximity value of a user profile of a user with that of the first user is a fifth function accounting for the number of exchanges between the two users. Advantageously, the proximity value of a user profile of a user with the first user is a sixth function comparing the actions performed by the two users on a predetermined set of documents and their similarities. Advantageously, the invention relates to a documentary search method comprising: a first step of defining a request; a second sorting step according to the method of the invention; a step of selecting and displaying at least one document whose relevance index is the highest of the list sorted in the second step.

The method of sorting a list of documents, each document comprising a semantic descriptor, takes advantage of the steps of defining and generating the semantic descriptors of each document. Advantageously, the invention relates to a system for implementing the method comprising: - first data storage means for recording and updating in a first database of the first set of users; a second storage means enabling the recording and updating in a second database of the user profiles; an interface making it possible to identify a first user; - Calculation means for calculating the proximity values between the users of the first set and the first user and the selection of the second set of users. Advantageously, the invention relates to a computer program comprising instructions for implementing the steps of the method. A technical effect of the invention lies in the efficiency of the data processing relating to the automatic and dynamic sorting of documents in digital format, and in particular to the inclusion in this processing of criteria that can be applied to elements of profiles, regularly updated, of users having access to these documents. Other features and advantages of the invention will be set forth in a detailed description & illustrated by the following figures: with the aid of the description which follows, given with reference to the appended drawings which represent: FIG. principle of a documentary research method according to the prior art; FIG. 2: the representation of steps of the method of calculating indices of interest of a list of documents according to the invention; Figure 3: a block diagram of the method of the invention; FIG. 4: a diagram of the means making it possible to implement the method of the invention; FIG. 5: a block diagram representing the steps of generating a semantic descriptor of the invention. Figure 2 shows a diagram comprising the main steps of the method of the invention. Definition of users A first user Up defines an R request through a user interface. A search engine MR, comprising calculation means, makes it possible to extract from a first database BDp a list of documents that are relevant to the predefined query. A request can be either explicitly defined when it is formulated by a user or it can be defined implicitly by a context defined by actions of the first user. Actions are only taken into account when the first user is identified by means of an interface. For example, a context may be the consultation of a given document from the document base. A request is then automatically generated from the context and recommendations are proposed to the first user as detailed in the following description.

A closed user network is defined, for example, by all users identifying with the user interface. It can be a particular community such as a corporate network or a social network of users sharing predefined data. The set of users of the network considered in the method of the invention is called the first set, it is denoted by U. We denote uke [1; N] an indexed user in a network comprising a number N of persons or users. The set of identified users of the network is the first set Ul. The users of the set Ul are registered in a first database of users, denoted BDu. The method of the invention comprises a first step 20 of determining a first user up of the first set from an interface. In one embodiment of the invention, an authentication or identification mechanism makes it possible to determine a first user up of the first set Ul. During the configuration step, the first user up can also define a profile or a user. number to connect to a network of users. The method of the invention makes it possible to identify a first user up among a network comprising uke [1; N] users. This first user up is defined as the user defining, explicitly or implicitly, a request R at a time To. Definition of the document base The method of calculating a social relevance index of a first document concerns a first document Dk given extract of the first documentary database, noted BDp. The BDp documentary database comprises a set of documents which constitutes for each of them an information medium. Each document in the document base includes a semantic descriptor.

A semantic descriptor describes the topics to which it refers and the semantic content of the document, including: - simple or compound terms; - simple sentences such as subjects, verbs, complements. Themes are weighted according to the importance they have in the document. Terms and sentences are weighted to be the most discriminating semantically. Terms and phrases that bring semantically closest documents closer together are used preferentially.

A semantic descriptor of the method of the invention may be developed by a given user or automatically generated from a semantic descriptor generation method, such as a semantic indexing engine mentioned above. It can also be created by a combination of these two methods. Its updates can either be done manually or automatically from a method of updating a semantic descriptor from a pre-established configuration allowing to configure at which moments an indexing is done and to define the policy of extraction of keywords or terms generated from the text of the document. The update of the semantic descriptor can also be a combination of these last two modes of updates. In the remainder of the description: Note (Di) ie [1; N] all the documents of the documentary database denoted BDp. We denote (Ai) ie [1; N] all the actions that can be performed by a user. All the actions include at least the following types of actions: - creation of a document; - a measurable assessment of a document; Depending on the embodiments, the types of actions may also include: - a modification of a document; - a consultation of a document; - a download of a document; - a selection of a part of the document; a transmission of a document by the transmission or reception of this document by means of an e-mail for example; - referencing the document by associating at least one term with the document. Finally, a user's action type may also include an action resulting from a combination of the previous actions.

Defining user profiles and interest profiles Each user includes a user profile. According to the embodiments of the invention, a user profile comprises: a profile of interests comprising semantic data including key words, themes, terms or complex sentences, the interest profile may according to the embodiments be defined by a user or updated automatically from actions performed on a set of documents of the set (Di) ie [1; N] or a combination of the two operations; a list of the actions performed on a set of documents of the set U1 which have been consulted or created; a set of transmissions comprising transmissions and receptions of a set of documents of the set U1. In one embodiment of the invention, the interest profile is generated or updated from a function making it possible to extract a list of terms from the semantic descriptors of the documents on which a user has performed at least one action. of the first set of actions and weight each extraction of terms by the type of action on each document.

In one embodiment of the invention, the number of transmissions of a document between two users or between a given user and a group of users accounts for common interests between the users. The interest is measurable by a posting of the items from an electronic mail and can in an alternative embodiment weight the interest according to whether a user has a profile "issuer" of documents or a profile of "receiver". That is, a user receives more often than he emits documents or vice versa. In one embodiment, the user profiles are created and saved in a second database denoted BDp. In other modes, user profiles can be created, saved, and updated in a file or index or table in another database. In addition, a third so-called social database, denoted BDSo, includes interactions between the documents and the users by saving the actions performed by each user on each of the documents of the BDp database.

User Profile Baking The process updates user user profiles over time and automatically updates user profiles according to the actions performed by all users. The updates of the user profiles can be performed on the BDP database including the recorded data of the profiles of interest as well as the other data stored in a user profile. The database BDP allows in one embodiment of the invention to record for each user u; the data of the semantic descriptors of each document, in particular the keywords, the themes or the complex sentences generated, corresponding to the documents on which actions have been carried out. The method of the invention makes it possible to take into consideration in the user profile the messages transmitted with another user containing for example a given document Dk. The themes of a semantic descriptor of a given document include semantic topics. A theme is not necessarily present in a document but it includes for example a given lexical field or a domain of a document. A profile of interest is the set of semantic data present in a user profile that are either user-defined or automatically collected from a user's actions on a document or a combination of both.

As a result, an interest profile can include collected themes of semantic descriptors of documents consulted. These themes can be, for example, considered as terms including a specific weighting. Several embodiments of the update of a user profile are possible: a first mode allows a regular update at predetermined times, for example at a defined time of a day, the update being done daily ; in this case the themes, the terms and the proximity values are calculated for each user at that moment; a second mode makes it possible to update the user profile of a user as soon as an action on a document is performed, in which case the complex terms and sentences and data that make it possible to generate the proximity values can be updated automatically in the user profile of each user. Updates to user profiles can also be done in real time or periodically over larger periods such as weekly or monthly. Example of defining user profiles Each user user profile includes a set of data to calculate a distance with the first identified user. The proximity value may be zero if a user of the first set Ul has no common interest with the first user up. This is for example the case if no message has been exchanged between the two users and if no theme or term in common has been collected in their user profile. In one embodiment of the invention, the definition of a user profile of a user can be enriched by other other data, for example: the definition of a first set of data, called data; explicit, of which a value may be calculated to represent a type and the nature of a task performed and / or to correspond to a function or position; 25 ^ a regular update of a second set of data, called implicit data, of which at least one value corresponds to a quantization of a number of data exchanges or messages transmitted comprising a given document with at least one other user; A third set of data, called semantic data including a list of terms, themes or complex sentences and possibly a recurrence index. In an exemplary embodiment, a task type may be determined from a list of prerecorded tasks. Typically an example of task 10 is "working on a project X". This task can be associated with a first value 15. The first digit 1 means "to work" or not and the second digit makes it possible to define a task among a number of predefined tasks, this is the task n ° 5.

In other equivalent examples, the method of the invention allows to assign a value to a predefined task. Another value may be assigned to the function or item more generally to a status of a user among the network. The method evaluates the status of a user from this value.

For example, in a hierarchical organization, the hierarchy level of a user can be associated with a value. The calculation of the distance between two users can take into account this last value. When groups are predefined in the user network, a group membership relationship can be measurable according to whether two users belong to the same group or to distant or close groups. When the definition of a user profile makes it possible to assign a list of terms from the actions on documents by a user in his profile of interest, these semantic data data are recorded. In an exemplary embodiment, each document includes a semantic descriptor. According to the embodiments, the semantic descriptor of a document can be generated automatically from the recognition of business terms of the document or by an action of a person collecting terms known as "representative" of the subjects addressed by a document or considered as generic to represent the topics covered in the document. The semantic descriptor of a document can also be generated by a combination of a manual action and an automatic referencing of the terms of the document. The terms also make it possible to reference the document in the document base. In one variant, the semantic descriptor can be considered as the first page of the document. The profile of interest of a user can be enriched by the recording of terms or themes collected in the semantic descriptors according to the type of actions performed by a user on said documents. The terms identified may be weighted by the type of actions previously defined on one or more documents.

A set of terms of a given user defined in his profile of interest can be for example: {bakery, bread, croissant, motorcycle, brake, tire}. When these terms are weighted by their recurrence in the actioned documents, the set of terms in the preceding example can be defined as: {bakery-2-4, bread-3-7, croissant -1-8, moto-5-3, brake-1-1, tire-2-61. The first digit associated with each term represents the number of occurrences of the term in the documents and the second digit associated with each term is used to weight the term by a value representing the type of actions. Thus, in this example, there are two occurrences of the word bakery, 3 occurrences for the word bread, and so on. The definition of the profile of interest of a user comprises, in one embodiment, the definition of themes. Each theme is associated with terms from the same family.

The method of the invention makes it possible to generate two themes from the terms of the list of the preceding example. For example, the first theme is: CULINARY PREPARATION, the second theme is AUTOMOTIVE MECHANICS. The method includes identifying a term and associating the term with a predefined theme. A score can be associated with the theme, taking the previous example, we count three terms of the first theme and three terms associated with the second theme. The scores of the topics can be weighted by the number of occurrences of terms relating to a predetermined theme and by the note concerning the types of actions of a user on a document. The method of the invention includes querying a table or a semantic file or database in which the complex terms or phrases associated with the themes are recorded. The method of the invention thus makes it possible to deduce the themes of the complex terms and sentences from the documents that have been the subject of actions by a user. The invention comprises several embodiments of the calculation of themes from the terms of a document.

In a first embodiment, from the occurrences of the terms in the documents of a predetermined corpus, the method calculates for each pair of terms a correlation coefficient. These correlation coefficients make it possible to construct a weighted graph. The most strongly connected components of this graph form sets of terms that are semantically related: they are themes. In a second embodiment, the method of the invention makes it possible to save the themes in a table, a register or a file or in the database. For each theme, a list of documents is associated with the theme. In one embodiment, a weighting is calculated according to the number of documents comprising themes corresponding to a given theme. The method makes it possible to calculate the most discriminating terms for each theme. The calculation of the topics includes the determination of the presence of terms in the documents and the number of occurrences of each term. In order to improve this calculation, in one embodiment, the system uses a method for determining semantic knowledge on the domain concerned. Several criteria can be used such as taxonomy, thesaurus or ontology.

The generation of themes may include the addition of terms that are not present in a document but whose document addresses the theme. For example, the method of the invention allows to associate broad terms interleaving less broad terms. An example is the term "programming" which is a theme including the terms "object language" and "JAVA". The theme "object language" is itself a theme including the term JAVA. The method of the invention makes it possible to weight the terms taken into account in each theme according to different criteria: the specificity of the terms, the multiplicity of occurrences, certain grammatical categories such as the named entities.

Definition of distance calculation / proximity value When considering a first user up, the method of the invention makes it possible to calculate a proximity value with each user of the first set.

FIG. 2 represents a step 21 of the method making it possible to determine a subset, denoted second set U2, of users of the first set which comprise a value of proximity greater than a predetermined threshold. The method of the invention makes it possible to define a second set of users included in the first set for defining a restricted number of users having a proximity of interest with the first identified user up who implicitly or explicitly performs a document search. A proximity value between two users can be calculated according to measurable distance criteria from measurable data of the user profile of each user. Different methods can be used to generate a second set U2 of users close to the first user up. One possible embodiment is one based on collaborative filtering techniques. This technique makes it possible to calculate the distance between two users by using, for example, the Pearson correlation coefficient. This technique makes it possible to calculate the distance between two users according to their actions on a set of documents and / or according to the data contained in each user profile.

To calculate the proximity value between two users of the first set, the method of the invention makes it possible to define a function comparing values generated by the data of two user profiles making it possible to calculate a distance between two users. A first distance can be determined by a function determining a first value calculated from a correlation between the terms of the interest profile, noted Tub of a user u; of the set Ul with the profile of interests, noted Tm, of the first user up. This first distance can take into account in one embodiment the number of occurrences of each term referenced several times in the profile of interests Tu; of a user u;, so as to weight the proximity value. A second distance can be determined by a function comparing a status of a user u; with the first user up, each user including a status, the different statuses being differentiated by a value representing a degree of difference. A third distance can be determined by a function counting the number of exchanges, for example of an electronic mail, between two given users or a given user such as the first user with a group of users. A fourth distance can be determined by a function comparing the actions performed by two users taken two by two or between a given user and a group of users on a predetermined set of documents and their similarities.

Finally, a distance can be determined by a weighting of the previous distances. The method of the invention makes it possible, from a document identified or consulted by a first user, to generate a second set U2 of the set U1 of users. The second set U2 is determined from a function making it possible to calculate a distance between each user of the set U1 and the first user and to compare it with a predetermined threshold. An advantage of the creation of this second subassembly is to reduce the costs and calculation times and to propose a list of documents relevant to the first user. The threshold may be data that is configurable and / or dependent on the user's environment. And from his user profile. It can be calculated for the second set U2 of users. For example, the 100 shortest distances between all users of the set Ul and the first user defines a variable threshold. An advantage of using a parameterizable threshold is that the method of the invention can adapt and configure itself adequately to different types of networks. For example, the threshold can be adapted to a network of users of greater or lesser size, from a few tens of users to several thousand or even hundreds of thousands for large networks. In addition, another advantage is to adapt to the disparity of subjects and thus themes of a documentary base. In particular, during step 20 making it possible to identify a user performing a search from a request, the method of the invention makes it possible to define a second set U2 of users having a value of proximity greater than a predefined threshold. . The method then makes it possible to identify themes that are of common interest for a set of users. The method of the invention makes it possible to define, in an alternative embodiment, groups of users according to certain common themes. In the latter case, the user profile of a user includes the identification of the group or groups of memberships of a user and in this case the method associates a value that can be compared. FIG. 2 represents a step 21 corresponding to the determination of the second set U2 of users having a proximity value greater than a predefined threshold with the first user. The calculation of the proximity value is a correlation function between the user profile of the first user and each user profile of the other users.

Definition of the social relevance index For a given document, denoted DK, and for the first user up, the method of the invention makes it possible to determine the value of a social relevance index, denoted Iso in a general manner and Ik in Figure 2. At a given time, a calculator allows to count, the set of actions performed on the DK document by the users of the second set U2 whose proximity value with the first user is greater than a predefined threshold. The method makes it possible to generate a score for all the actions performed by the users of the second set U2 of users on the document DK. FIG. 2 represents a step 23 in which is computed the social relevance index Ik of a document DK for the second set U2 of users calculated in step 21 having a proximity of interest with the first user.

Thus, when the method applies to a set of documents (Di) ie [1; N], the documents can be sorted in ascending or descending order of the scores calculated by the method of the invention of each document Di.

The invention relates in this sense to a method of sorting a predetermined list of documents and can therefore be combined with a first search engine returning a list of documents from any conventional search engine. The sorting method of the invention makes it possible to sort an ordered list generated previously by a semantic search engine. An advantage of the sorting method of the invention is that it is compatible with any search engine since it can be applied to any ordered list of documents. FIG. 2 represents a selection step 22 that makes it possible to sort a list of relevant documents of which at least one action has been performed by the users of the second set U2, that is to say those having the smallest distances with the first user. In Figure 2, the actions are denoted by (Ai) ie [1; n]. The method of the invention comprises a step 23 for defining for a given document Dk and a given group of users a relevance index of the document Dk. One embodiment of the invention makes it possible to exploit a so-called social database denoted BPso which comprises for each document of the database BDp, the users having performed at least one action on a document and the type of actions. Optionally, the database lists the occurrences of actions performed by users on documents. Step 23 of the method of the invention of FIG. 2 makes it possible to interrogate one or more databases making it possible to render the type of actions of a user or of a set of users on a given document Dk of the BDp database. In one embodiment, a database stores for each pair {document Dk, user ui} a value that is changed according to the type and number of actions performed on the document Dk over time. This value corresponds to the interest the user has in document Dk.35 An example of calculation of the social relevance index For each of these actions, a value is attributed which can be weighted according to the number of identical actions performed. This is a score that measures the interest of a document by a set of users.

An example allows to consider a user Uq who would have performed the following actions: - Creation of a document D1: value 10 - Creation of a document D2: value 10 - Download of a document D3: value 100 - Selection of a document change from document D3: value 100 - Modification of the document D3: value 10 - Appreciation of a document D4 by a positive score: value 1 - Appreciation of a document D5 by a positive score: value 1 Each document includes an index corresponding to a first value V1 which is calculated according to the type and number of actions performed on the document over time. The value V1 is the user's interest in this document. The first value V1 can be increased or decreased depending on whether the actions are configured and their number to be taken into account positively or negatively.

The first value V1 includes, for example, the previous values for defining the interest of a document by a set of users. In the previous example, the first value V1 can be defined as the sum of the previous values. The method of the invention thus makes it possible to define a value V1 = 232. Each unit specifies a degree of appreciation. For example, the number 2 of the units represents a "notion of collaborative participation", the digit 3 of the tens represents the "degree of expertise" of a user when he modifies or creates a document, the number 2 of the hundreds represents the quantification of a "knowledge sharing". Other equivalent ways of defining a first value V1 can be defined. The first value V1 makes it possible to define a qualification of the global interest of a document by a set of users.

In the same way, a given document Dk may have been the subject of identical or different actions of a second set U2 of users. The community of users of the second set U2 having performed an action on a given document is taken into account in the evaluation of the index of social relevance ISo. U2 is established from the calculation of a distance between the first user and the users of the first set. A predefined threshold makes it possible to select a subset of users having a common interest with the first user. In this case, another method of calculating a second value makes it possible to take into account the number of actions performed by a set of users of the second set U2 and the type of action performed by each one of them. Let's take two examples where DK = DI and Dk = D2. The second value V2 represents the social life of a document.

This value depends on the document. An example for a document D1 was: - "liked" by 16 people of the second set U2, - "consulted" by 20 people of the second set U2, - "not appreciated" by 2 people of the second set U2, - "registered" by 10 people from the second set U2.

The second value V2 which represents the social life of the document D1 is equal to 15 for example. In this example, let us take a second document D2 which was: - "loved" by 1 person of the second set U2, - "consulted" by 20 people of the second set U2, ^ "not appreciated" by 12 people of the second set U2, - "registered" by 0 people from the second set U2. Thus, the second value V2 of D2 is equal for example to 3. This implies that the document D1 has a social life more "interesting" than D2 and therefore it is more relevant "socially" than D1 for the first user. In one embodiment it is possible to take into account a third value which may be a combination of the first and second values or to choose an equivalent calculation mode for calculating a social relevance index ISo of a given document Dk.

Another example makes it possible to define the first value representing the level of interest for a document, in particular by a weighting by the type of action of a unit value. Thus a document creation may reflect a user's interest twice as much as a download of the same document and two times less than a modification of the same document. The method of the invention makes it possible to assign a note to the interest in a given document by a set of users. The method of the invention also applies to a recommendation mechanism. In the latter case the first user up is a user who is connected to the network comprising the first set Ul of users. The recommendation is made by submitting a list of proposed documents to the first user during actions taken by the first user up. Typically, when viewing a document or a data carrier, some relevant documents are offered to the first user. In place of the request formulated by the first user through the semantic search engine, the semantic descriptor of a document accessed by the first user makes it possible to define an enriched request to the engine. The set of data, the semantic descriptor or a part filtered according to a given context of the data of the semantic descriptor, makes it possible to generate according to the method of the invention a list of documents to the first user interactively.

When the first user makes a request in a semantic search engine, the method of the invention is implemented by an active approach of the first user. When the first user consults a document, the method of the invention is implemented by a passive approach of the first user. The set comprising a given user profile, a given action on a document and the semantic descriptor of the document makes it possible to define a context. The definition of the context makes it possible to initiate the recommendation, that is to say the submission of a list of documents likely to interest the user.

In the case of a recommendation, the step of defining a request is not necessary. But steps 22, 21 and 23 of the process of the invention are carried out. FIG. 2 thus represents a first up user identifying with an interface 20 and formulating a request R in a search engine MR. A semantic search engine 24 generates a first list of documents from a request R of a corpus of documents D and a semantic indexing engine S. A function denoted F4 (R, D) makes it possible to generate a score for each document of a document base making it possible to establish this first ordered list in ascending or descending order of the scores K of each document. A step 21 of the method makes it possible to define a second set U2 of users from a function d = F1 (up, u;) making it possible to calculate a value of proximity between the users of the first set U1 and the first user. For each document DK of the first list, a social relevance index Ik is calculated from a function F2 (u ;, Dk) and the second set U2 taking into account at least all of the actions A; performed by each user u; of the second set U2 on said document DK. A step 22 makes it possible to generate a sorting of the first document list so as to generate a second list of documents ordered in ascending or descending order of the calculated social relevance index of each document of the first list.

In an alternative embodiment an index of interest is calculated from the score and the social relevance index of each document according to a function rated PERTp. This index of interest can be weighted by a coefficient allowing more or less privileging the score than the index of social relevance or vice versa.

Finally, in this variant embodiment, a function denoted ORD = F6 (PERTp, p) generates a sorted list of documents according to the index of interest. FIG. 3 schematically represents the generation of a first list LI sorted from a semantic search engine MR according to a score of each document of the documentary database and a list according to a list L2 sorted and filtered from an engine. collaborative search method Fc corresponding to steps 21, 22 and 23 of FIG. 2. A user up identifies himself from an interface denoted VISU which makes it possible in particular to generate, in one embodiment, requests R comprising semantic data . Queries in another mode can be generated automatically from the actions of a user, this is the recommendation defined previously. FIG. 4 represents the means making it possible to implement the invention. In particular, a possible architecture makes it possible to have a first database BDp storing the documents shared by the first set Ul of users. In one embodiment, a possible configuration for implementing the method of the invention is the creation of several databases for storing the data of the system. Other equivalent configurations, particularly as regards the grouping of data in the same database is a possible alternative to implement the method of the invention. A possible configuration includes the definition of: - A first database to save, edit, create and delete users, it is denoted BDu; - A second database to save, edit, create and delete user profiles of users, it is denoted BDp; - A third database, called social, to save, edit, create and delete the actions performed by each user on each document in the database, it is noted BDso; - A fourth database to save, edit, create and delete distances between users of the database, it is noted BDDIsT. - A fifth database to save, edit, create and delete the semantic descriptors of each document in the database BDD, it is denoted BDps. In another embodiment the first BDu and second BDp databases can be grouped into a single database.

Similarly, in another embodiment, the fourth database BDDIsT can be replaced by a calculation for each new operation of a user from a new distance calculation. Action types, intrinsically to users and documents, can be stored either in a database or in an index or file. In an alternative and equivalent way, databases can be replaced by indexes, files or tables of a more general database. Various data organization means may allow the implementation of the method of the invention. The method of the invention is also implemented by a WEB server, denoted S-WEB, comprising at least a motherboard, a processor, a memory, a hard disk and a network interface. A WEB client for each user provides access to the databases and the web server. The WEB server also includes an operating system. The web server comprises in one embodiment of the method of the invention: a web interface allowing access to the applications and data of the server; an authentication server for identifying a user connecting to the network; a document manager allowing an interface with the first database BDp so as to allow the management of documents including the addition, the modification, the deletion, and all the authorized actions on the documents; a user manager allowing access to the user database and the management of the users, in particular their addition and deletion; - A profile manager that allows the creation of a user profile, its updating by the user and its automatic update and its deletion; a theme calculator allowing from a document parser to generate themes from terms of each document; a social network making it possible to administer and manage links between users, in particular on home groups, common tasks; - a message manager to know the quantization of messages received and sent from a user with other users and develop associated scores; a user action manager for recording each action of a user on each document, as well as the type of actions performed; a semantic indexing engine making it possible to interface with the fifth database BDps which makes it possible in particular to assign a list of terms, from themes of complex sentences to a document in its semantic descriptor; exchanges interfaces with the fifth database BDps and the first database BDu and the first database BDp so as to record and extract data in the databases from requests made on the web server. More generally, the Web server designates both a computer that acts as a computer server that hosts an Apache-type http server software or an Apache-type server program itself and another server program that makes it possible to communicate. with other servers, managers and calculators In one embodiment, the message manager can be replaced in an equivalent manner by a microblogging manager to know the interest of a user in messages sent by another user and to develop associated scores. In another variant embodiment, alternatively, a message traffic record server makes it possible to receive from a client of message traffic records the quantization of the messages received and sent from a user with a message. other users and develop associated scores. Conversely, a message traffic record client is used to send to the message traffic record server the quantization of messages received and sent from a user with other users. Finally, the method of the invention is implemented from a WEB client, denoted C-WEB, also called a user interface that makes it possible to exchange data with the WEB S-WEB server through an N network. C-WEB client includes at least one motherboard, a processor, a memory, a hard disk and a network interface so as to connect each WEB client to the rest of the network.

An operating system and web browser allow to offer a user an enriched interface. The invention relates to a computer program comprising instructions for carrying out the steps of the method of the invention. In particular, the computer program includes the instructions so as to generate queries for extracting information from the semantic descriptors of documents from the document base and to return bibliographic data of each document in the form of a list. The computer program is run from computing means on a data server. The instructions are generated from queries from so-called "client" computers from which users connect. A client interface is used to define instructions to the computer program so that it executes steps of the method.

FIG. 5 represents the steps of generating a semantic descriptor DS of a document Dk comprising data of which: a first set, denoted EdMC, of data representing the key words generated from the semantic data of at least one document Dk; a second set, denoted EdTP, of data representing the parent terms, called concepts in the remainder of the description, semantic data of a document Dk; a third set of data, denoted EdT, representing the generic relations of triplets of terms generated from triplet relation of a document Dk.

The data in the EdMC set are generated from a dictionary, denoted PEL, which includes lexical data. The parent terms are generated from a first set of concepts, denoted PEN, which represent data which is similar to a dictionary in which terms are represented in a graph making it possible to determine the parent terms of the terms of a document . The data of the EdTP set is generated from a set of predetermined generic relationships, denoted PER. A pointer makes it possible to determine the meaning of a relation of a triplet of terms.

Optionally, a temporary index is generated, denoted by IT, making it possible to select the terms and to generate intermediate or temporary lists. This step is optional. The generation of the semantic descriptor DS includes steps for collecting relevant data to facilitate the execution of the sorting method of the invention. In particular, so as to generate semantic descriptors for each document of the document base, the method comprises a definition step: of at least a first lexical set of terms comprising the canonical form and the grammatical nature of each term; o at least a first set of concepts, each concept representing a node of a graph, each concept having one or more kinship relations, each node having a predefined degree of kinship, each concept having one or more relations with a term of the first lexical set of terms; o at least the definition of a first set of conceptual relations, each conceptual relation representing a node of the graph, each conceptual relation being able to have with other conceptual relations zero, one or more relations of kinship; o at least the definition of a first set of conceptual facts, each conceptual fact being defined by a conceptual relation of the first set of conceptual relations, each conceptual fact associating two concepts of the first set of conceptual relations; concepts so as to form at least one ordered triplet. In addition, the generation of the semantic descriptor DS comprises: a step of identifying a first document; a step of generating a temporary index comprising each term of the first document common to the first lexical set; o a step of generating a set of key words of the first lexical set present in the temporary index; a step of generating at least one set of concepts of the first set of concepts from the analysis of: the terms of the temporary index; relations between the terms of the first lexical set and the first set of concepts; ^ kinship relations between concepts; a fourth step of generating a second set of conceptual facts from the correlation of the analysis of the terms of the temporary index; relations between the terms of the first lexical set and the conceptual relations of the first set of conceptual relations; relations of kinship between conceptual relations; o a fifth step of generating a set of concepts chosen from the correlation of the analysis: ^ of the concepts generated in the third step ^ of the relationship between the concepts ^ of the number of occurrences of terms related to these concepts by relationship; a last step of generating a semantic descriptor comprising: the set of key words; ^ the set of concepts chosen in the fifth step; ^ the set of conceptual facts generated in the fifth step.

Finally, we detail an exemplary embodiment for illustrating an implementation of the method for generating a semantic descriptor DS. The document is considered a signal. The initial signal is a list of characters, denoted LCr, encoded in UTF-8 format. Each processing step enriches this initial signal with information. We note Tokenization as the first step to generate keywords. The tokenization creates a token list, denoted LTk, from LCr.

Token type definition: A token is a sequence of characters that do not contain spaces. There are three types of token according to the Unicode character classes: - a "word" token TTkWord is a sequence of characters beginning with a letter and containing only letters and numbers; - a number-type token TTkNumber is a sequence of characters beginning with a digit and containing only numbers; - a "special character" token TTkSpec is a character that is neither a letter nor a number nor a space. Adjacency Definition: Two tokens are adjacent if they are not separated by one or more "space" characters. A token is characterized by: - its type, taken from the values TTkWord, TTkNumber and TTkSpec; a "start offset" and an "end offset" which determine the LCR sub-list to which it relates; - its adjacency, a "true" or "false" value in relation to the previous token and the next token.

A list of LTk tokens is thus deterministically produced.

Several types of detections can be associated to process a document to generate keywords, also called "tags". The following steps are numbered 1 to N and sub-steps i.1 to i.M for easy reading. No sequential order is imposed for the execution of the method by the numbering, the steps can be performed independently of each other. Step 1: Detection of simple terms A simple term is a sequence of tokens that forms a term and links it to a concept by a relation "has for meaning".

To detect terms and link them to concepts, we use: - dictionaries that associate each term with a concept; - Token automatons that create terms and associate them with a concept. A dictionary links each term to one or more concepts by a relation "for meaning": - France, Spain: type = Country - Paris, Lyon, Marseille: type = City - Alphonse, Béatrice: type = First name - Durant, Dupond: type = Name - Martin: type = First name, Last name - Albert Einstein, Jean Jaures: type = Celebrity - UN, UNICEF: Organization - Google, Microsoft: Company - BBC, France Culture: Media - default, fixed rate loan: BetBank A token automaton recognizes one or more of the following characteristics: - the type of the token; the character sequence delimited by the start offset and the end offset; - the adjacency between the token. An automaton is defined by a constraint pattern on the token it seeks to recognize. A constraint, described between a pair of braces, specifies the values that the attributes of a token can take: - type: one or more possible types; if not specified, all types are accepted; - seq: acceptable character pattern as a regular expression; if not specified, all strings are accepted; - adjNext: can take two values, "true" or "false"; the value is true if and only if the token must be adjacent to the next token; if not specified, the adjacency of the tokens is not constrained.

The PLC definition also contains the concept recognized by the PLC. Step 1.1: Unit Detection Examples: ° C A unit is recognized from a list of all units described as PLCs.

List: - type = TTkSpec, seq = "0", adjNext = true type = TTkWord, seq = "C"}

Concept = "unit" Step 1.2: Detection of ordinal adverbs Examples: "2O1" 301 ... PLC: {type = TTkNumber, adjNext = true} {type = TTkSpec, seq = "°"}. Concept = "adverb.latin" Step 1.3: Detection of ordinal abbreviations Examples: "2nd", "3rd", ... Automaton: {type = TTkNumber, adjNext = true} {type = TTkSpec, seq = {"e", "es", "er", "ers", "re", "res", "ièmes" or "èmes"}}. Concept = "abbreviation.ordinal" Step 1.4: Detection of editorial abbreviations Values: "n °" "d °" "v °" "r °" "f °" Automaton: {type = TTkWord, seq = {"n °" , "d °", "v °", "r °", "f °"}, adjNext = true} {type = TTkSpec, seq = 101}. Concept = "abbreviation.edition" Step 1.5: Fraction detection Examples: "1/2", "3/4", but also "A" or "Y4". PLC: {type = TTkNumber} {type = TTkSpec, seq = {type = TTkNumber} 1 {type = TTkSpec, seq = {"'A" or "¾"}} Concept = "math.expression.fraction" Step 1.6: Detection of compound words with hyphen Examples: "sleeping car", "vis-à-vis". Controller: {type = TTkWord} ({type = TTkSpec, seq = "-"} {type = TTkWord}) * Concept = "name.composed" Step 1.7: Detection of enclitics Examples: "myself", "themselves same "," self "," ourselves "," yourself "," do I "," do you "," do it "," do it "," do we " , "do you", "do they", "do they", "do it", "do it", "do it", "do it", "do it", "do it" is this ", etc.? Automaton: {type = TTkWord, seq = {"me", "she", "they", "them", "self", "us", "you"}} {type = TTkSpec, seq = "-"} {type = TTkWord} It {type = TTkWord} {type = TTkSpec, seq = "-"} {type = TTkWord, seq = {"I", "you", "me", "you", "it", "they", "she", "they", "we", "you", "the", "the", "in", "y", "this"}} Step 1.8: Ephelcystic Detection Examples: " he said, "go away", "give me some" Automate: {type = TTkWord} {type = TTkSpec, seq = "-"} {type = TTkWord, seq = "t "} {type = TTkSpec, seq =" - "} {type = TTkWord} It {type = TTkWord} {type = TTkSpec, seq =" - "} {type = TTkWord, seq = {" t "," m " }} {type = TTkSpec, seq = "'"}} {type = TTkWord} Concept = "ephelcystic" Step 1.9: Detection of the' - 'used as a parenthesis Example: "Since to follow - and see to be denied - the military principles of Saint-Loup, I had followed with great detail the Boer War, I had been led to reread ancient stories explorations, travels. " An alternative typography is to use the character - without spaces before and after. Automatic: {type = TTkWord, adjNext = false} {type = TTkSpec, seq = "-"} {type = TTkWord} It {type = TTkWord, adjNext = false} {type = TTkSpec, seq = "-"} {type = TTkWord} There {type = TTkWord} {type = TTkSpec, seq = "-", adjNext = false} {type = TTkWord} Concept = "segmentation.bloc" Step 1.10: Detection of literal numbers with or without hyphen Example: entities of the form "one hundred and twenty-three" or "one hundred and twenty-three" (modern spelling) or "one hundred and twenty-three" (bad or old spelling). Let numberSet = = {"one", "two", "three", "four", "five", "six", "seven", "eight", "nine", "ten", "eleven", "twelve" "thirteen", "fourteen", "fifteen", "sixteen", "twenty", "thirty", "forty", "fifty", "sixty", "hundred", "thousand", "million", "million", "billion", "billion", "trillion", "billion", "trillion", "trillions"} Automaton: {type = TTkWord, seq = numberSet} ({type = TTkSpec, seq = "- "}? {type = TTkWord, seq = numberSet}) Concept =" general.literalNumber "Step 1.11: Acronym detection Examples: Simple acronyms with '.' : "SNCF" (the new form is "SNCF"), but also "complex" acronyms such as: "SUD-PTT", "CFDT-Fgaac", "SUD-E", etc. PLC: {type = TTkWord, seq = {AZ}} ({type = TTkSpec, seq = "."} {Type = TTkWord, seq = {AZ}}) * It {type = TTkWord, seq = {A-Zr} {type = TTkSpec , seq = "-"} {type = TTkWord, seq = {AZ} {a-zr} Concept = "general.acronym" Step 1.12: Detection of abbreviations of simple or compound first names Examples: - First names: "F", "A", "Ch", ... - Well-formed first names: "F.-Y.", "J.-J.", ... - Poorly formed first names: "FY", "M." , "FY", "JJ", ... Let abbPrenSet = {AZ, "BI", "Br", "Ch", "Cl", "Br", "FI", "Fr", "Gh", "GI", "Gr", "Kh", "KI", "Ph", "Pr", "Sh", "Th", "Tr", "VI", "Wh", "Chr"} PLC: {type = TTkWord, seq = abbPrenSet} {type = TTkSpec, seq = "."} There ({type = TTkSpec, seq = "-"} {type = TTkWord, seq = abbPrenSet} {type = TTkSpec, seq = " . "}) * Concept =" familyName.abbreviation.firstNa me "Step 1.13: Detection of punctuation Examples: - The separations of blocks within a sentence: two points, comma, double quotte, parenthesis or hook, '..', '...' - Sentence separators unambiguous: semicolon, question mark or exclamation mark, asterisk. Step 1.14: Detection of integers and decimals Example: French and Anglo-Saxon formats are supported, but not the exponential scientific notation. PLC: {type = TTkNumber, adjNext = true} ({type = TTkSpec, seq _ "\ .I,"} {type = TTkNumber})? Concept = "general.number" Step 1.15: Detection of the apostrophe Example: Grouping of the apostrophe with the token on the left: "1" followed by "" 'becomes "I" ", etc. Same treatment for the following cases : The text portions "GWENC'HLAN" and "EVAN'S" produce the token lists grouped together "GWENC" '+ "HLAN" and "EVAN"' + "S".

Automaton: {type = TTkWord, adjNext = true} {type = TTkSpec, seq = "" '} Concept = "general.rightApostrophe" Step 1.16: Date Detection Example: 2010/10/08; 12-08-1971; July 14, 1789 We define: monthPattern = "januaryfbruarylmarslavril Imailjuinljuilletlasttobeptemberembelctober november" Automate: {type = TTkNumber, adjNext = true} {type = TTkSpec, seq = "-II / adjNext = true} {type = TTkNumber} {type = TTkSpec, seq = "-1 I / adjNext = true" {type = TTkNumber} Il {type = TTkNumber, adjNext = false} {type = TTkWord, seq = monthPattern, adjNext = false} {type = TTkNumber} Concept: "date" Step 2 : Creating complex terms and deleting simple terms An inference engine is used to create complex terms. The initial facts are the adjacency relations between the terms.

The rules have the effect of: - deleting simple terms to manage ambiguities; - create complex terms by concatenating several terms and linking it to a relation "has for meaning" to a given concept; - add "meaning" relationships between terms and concepts. Examples legislative texts: the recognized forms are: decree no 72-1263 of December 29th, 1972 (decree, law, Law, decree, Decree), law n ° 2005-380 of April 23rd, 2005, article 8 of the law of July 17th, 1975 . ({type = TTkWord, seq = "article", adjNext = false} {type = TTkNumber, adjNext = false} {type = TTkWord, seq = "(of laldu)", adjNext = false})? {type = TTkWord, seq = "(decree) (n °? Ino? ln adjSext = false} {type = TTkNumber, adjNext = false} ({type = TTkSpec, seq =" - ", adjNext = false} {type = TTkNumber})? ({Type = TTkWord, seq = "from", adjNext = false} {type = Date})? Concept = "right.citation.legislative" Step 3: recognition of relations In the same way as for concepts we can recognize conceptual relations, eg the term "leads" is related to "a for sense" with the conceptual relation "Drive." Step 4: Calculation of taqs, also called "keywords" This step consists of deciding which tags will be generated from the terms, based on the characteristics of the terms, their relation to the text (number of occurrences, presence in the title, etc.) and the recognized concept, a function determines whether the term should be retained or not. and determining its weight By iterating on all the terms, one builds a list LTag of ter Step 5: Detect Conceptual Facts Previous processing generated for each sentence an LCR list of detected concepts and detected relationships.

Example: the phrase "Michel Dupont drives his car. ". The term "car" is related by a relation "has for meaning" to the concept "[Car]", so the concept "[Car]" is present in the sentence. The term "Michel Dupont" is linked by a relation "has for sense" to the concept "[Person]". The term "leads" is linked by a relation "a for meaning" to the relation "(Conduct)". Definition of the conceptual relations One defines a set of conceptual relations ER. Some conceptual relationships have related conceptual relationships. Example: The relationship "Drive" has the parent relationship "Pilot". Definition of the signatures of the conceptual relations The signature of a conceptual relation is a pair of concepts (C1, C2). Example: The "Drive" relationship has the signature {"[Human]", "[Vehicle]"}. We have a set of conceptual relations RC, a relation of kinship between these conceptual relations, and signatures of conceptual relations. Definition of a conceptual fact: A conceptual fact is a triplet consisting of a concept, a relation, and a concept. Here is the method to detect the presence of these conceptual facts: - For each sentence of the text P, we consider the LCR list of the concepts of the sentence and the relations of P. - If the LCR list contains two concepts C1 and C2 and a relation R1 such that there exists a relation R2 of signature (C3, R2, C4) such that: R1 = R2 or R1 has for ancestor R2 o C1 = C3 or C1 has for ancestor C3 o C2 = C4 or C2 a for ancestor C4 - Then we add the conceptual fact (C1, R1, C2) to the list of CFL conceptual facts. Example: the phrase "Michel Dupont drives his car. ". The sentence contains two person concepts C1 = "[Person]" and C2 = "[Car]" and a relation R1 = "(Drive)". The relation R2 "(Check)" has a signature which is (C3 = [Person], R2 = (Control), C4 = [Vehicle]) such that: - R1 = "(Drive)" has ancestor R2 = "( Check) "- Cl =" [Person] "= C3 - C2 =" [Car] "has for ancestor C4 =" [Vehicle] "The conceptual fact" ([Person: Michel Dupont], (Check), [Vehicle] ) is added to the list of conceptual facts.

A semantic descriptor is defined by: - the list of selected terms LTag - the list of LFC conceptual facts The document base associates with each document d of D a semantic descriptor ds of DS.

A semantic document base BDS consists of a set (di) i of documents and associated descriptors (dsi) i generated as previously described from a dx document of D. The system calculates its DS dsx semantic descriptor. When carrying out a semantic search, for example by recommendation as defined previously in the description of the invention, the system then calculates for each semantic descriptor (dsi) i a similarity value sim (i) as a function: dsi LTagi terms and dsx LTagx terms - RTMi terminology relations and RTMx terminology relations Finally, here are some advantages of the method of the invention which make it possible to: - improve access to relevant information in companies regardless of their size. This implies a reduction of information search time; - value all the knowledge present in the various documents and the actions performed on them; to improve the sharing of information within an organization comprising a closed set of users such as a company; 25 30 - detect changes in skills of certain users of an organization. - Promote the exchange and sharing of information between users of a network - Create links between users implicitly: creation of user communities with the same interests.

Claims (20)

REVENDICATIONS1. A method for calculating a social relevance index (ISo) of a first digital document (Dk) accessible to a first set (U1) of users (ui), each user (ui) being capable of producing a first set of actions ((Ai) ie [1; N]) on the first document (Dk), characterized in that the method comprises: - a step of identifying a first user (up) of the first set (U1); a step of calculating at least one proximity value (dip) with the first user from a set of data included in a user profile (Pi) of each user (ui), a selection step in the first set (U1) of a second set (U2) of users whose proximity value is greater than a predetermined threshold (drain), each calculated proximity value being dependent on the first user; a step of calculating, from the calculator (K), the social relevance index (ISo) from the set (AkU2) of the actions performed by each user of the second set (U2) on the first document ( Dk). 2. Method for calculating a social relevance index according to claim 1, characterized in that the first set of actions ((Ai) ie [1; N]) comprises the modification of a document, the creation of a document and the consultation of a document. Method for calculating a social relevance index according to claim 1, characterized in that the first set of actions ((Ai) ie [1; N]) comprises the generation of a measurable assessment of a document . 4. Method for calculating a social relevance index according to claim 1, characterized in that the first set of actions ((Ai) ie [1; N]) comprises the transmission of a document by means of a email.35 5. A method of calculating a social relevance index according to claim 1, characterized in that the first set of actions "Honey; ni]) comprises a download of a document or a selection of a part of the document. 6. A method of sorting a list of documents, characterized in that it comprises: a step of calculating, from a calculator, a set of indices of social relevance (Iso) according to the method of claim 1, of a first list (LI) of documents ((Di) ie [1; mi) ordered according to a first order (ORDRE1); a step of sorting, from a calculator, the first list (LI) of documents ordered according to a second order (ORDER2) corresponding to a descending order of the value of each social relevance index (Iso). 7. A method of sorting a list of documents according to claim 6, each document (Di) comprising a semantic descriptor (DSi) referencing a set of semantic data (Tipi) specific to each document (Di), characterized in that it comprises: a step of defining a first request (RI) defined by means of a user interface, a user defining the request for determining a first user (up), said request (RI) comprising at least one term; (SORTING); a step of generating, from the computer (K), a second list (L2) of digital documents extracted from a first documentary database (BDD), the generation of the second list (L2) being triggered from of the first request (RI), the second list (L2) being ordered from a first function (FI) for generating a semantic relevance index attributing to each document (Di) a value (Vi) resulting from a correlation between the terms of the request (RI) and the semantic descriptor (DSi) of each document (D;), the second list (L2) being ordered in descending order of the value of the semantic relevance index, the second list (L2) including the first list (LI) of documents (D;). 25 30 8. Method according to one of the preceding claims, characterized in that a user profile (Pi) of a user (ui) comprises a list of actions (Ai) performed on each of the documents of the first documentary database (BDp). 9. Method according to one of the preceding claims, characterized in that a user profile (Pi) of a user (ui) comprises a set of transmission of the documents of the first database (BDp) and the recipients associated with the transmissions. . 10. Method according to one of the preceding claims, characterized in that a user profile (Pi) comprises a profile of interests (-fui) comprising semantic data including a list of terms. 11. Method according to claim 10, characterized in that the proximity value (dip) of a user profile (Pi) of a user (ui) of the second set (U2) with the first user (up) is calculated from from a second function (F2) determining a first value calculated from a correlation between the terms of the interest profile (-fui) of each user (ui) of the first set (U1) with the interest profile of the first user (Tm). 12.Procédé according to claim 11, characterized in that the list of terms referenced in each profile of interest (-fui) comprises the number of occurrences of each term referenced several times in the profile of interest (-fui) of each user (ui) of the first set (U1), the second function (F2) weighting the common terms between the interest profile (-fui) of each user of the first set (U1) with the interest profile (Tm) of the first user (up) by their number of occurrences, a second value being calculated. 13. Method according to any one of claims 9 to 12, characterized in that the interest profile of a user of the first set (U1) is generated from a third function for extracting a list of terms. of the semantic descriptor (DSi) of at least one document (Di) on which the user has performed an action of the first set of actions ((Ai) ie [1; N]) and weighting each extraction by the type d action on the document (Di). 14. Method according to one of the preceding claims, characterized in that the user profile comprises a fourth value corresponding to the result of a fourth function comparing a status of a user with the first user (up), each user (u;) including a status, the different statuses being differentiated by a value representing a degree of difference. 15. Method according to any one of claims 9 to 14, characterized in that the value of proximity (d; p) of a user profile (P;) of a user (u;) with that of the first user ( up) is a fifth function accounting for the number of exchanges between the two users. 16. A method according to any one of claims 8 to 14, characterized in that the proximity value (d; p) of a user profile (P;) of a user (u;) with the first user (up) ) is a sixth function comparing the actions performed by the two users on a predetermined set of documents and their similarities. 17. A documentary search method comprising: a first step of defining a request; a second sorting step according to one of the claims 6 to 16 of documents; a step of selecting and displaying at least one document whose relevance index is the highest of the list sorted in the second step. 18. A method of sorting a document list according to any one of claims 7 to 16, each document (D;) comprising a semantic descriptor (DS;), characterized in that the semantic descriptor (DS) is generated at from the following steps: a step of definition: of at least a first lexical set of terms comprising the canonical form and the grammatical nature of each term; o at least a first set of concepts, each concept representing a node of a graph, each concept having one or more kinship relations, each node having a predefined degree of kinship, each concept having one or more relations with a term of the first lexical set of terms, o at least the definition of a first set of conceptual relations, each conceptual relation representing a node of the graph, each conceptual relation having, with other conceptual relations zero, one or more kinship relations; at least the definition of a first set of conceptual relationship signatures, each conceptual relationship signature associating with each conceptual relation at least one concept pair of the first set of concepts; o at least the definition of a first set of conceptual facts, each conceptual fact being defined by a conceptual relation of the first set of conceptual relations, each conceptual fact associating two concepts of the first set of concepts so as to form at least one ordered triplet, - a step of generating the document descriptor, comprising: a step of identifying a first document; O a step of generating a temporary index comprising each term of the first document common to the first lexical set; o a step of generating a set of key words of the first lexical set present in the temporary index; A step of generating at least one set of concepts of the first set of concepts from the analysis of the terms of the temporary index; relations between the terms of the first lexical set and the first set of concepts; ^ kinship relations between concepts; a step of generating at least one set of relations of the first set of conceptual relations from the analysis of the terms of the temporary index; Relationships between the terms of the first lexical set and the first set of conceptual relationships; relations of kinship between conceptual relations; a fourth step of generating a second set of conceptual facts from the correlation of the analysis of the terms of the temporary index; relations between the terms of the first lexical set and the conceptual relations of the first set of conceptual relations; relations of kinship between conceptual relations; ^ signatures of conceptual relations; O a fifth step of generating a set of concepts selected from the correlation of the analysis: ^ of the concepts generated in the third step ^ of the relationship between the concepts ^ of the number of occurrences of the terms related to these 20 concepts by the relation; ^ the structure of the document in which this concept appears; ^ conceptual facts generated in the fourth stage in which the concept appears; A last step of generating a semantic descriptor comprising: the set of key words; ^ the set of concepts chosen in the fifth step; 30 ^ the set of conceptual facts generated in the fifth step. 19. A system for carrying out the method of any one of claims 1 to 18 comprising: first data storage means for recording and updating in a first database (BDu) ) of the first set of users (Ul); a second storage means for recording and updating in a second database (BDP) user profiles ((Pi) ie [1; N]); an interface making it possible to identify a first user (up); - Calculation means for calculating the proximity values (dip) between the users of the first set and the first user and the selection of the second set of users (U2). 20. Computer program comprising instructions for carrying out the steps of the method according to any one of claims 1 to 16.