FR3059797A1 - AUTOMATED METHOD OF ESTABLISHING THESAURUS OF NAMED ENTITIES THAT CAN INCLUDE A PLURALITY OF HIERARCHICAL LEVELS, AND THE USE OF SUCH THESAURUS - Google Patents

Abstract

A method for automatically generating text data from a named entity thesaurus, wherein a named entity may include a plurality of entities of different hierarchical levels, including: an extraction step, from text data, of a character string designating a named entity; a string processing step for establishing a list comprising at least one entity segment; a step of eliminating the list of entity segments already present in the thesaurus, to form a list of new segments; and a step of updating the thesaurus for each new segment of the list of new segments. The invention also relates to a method for automatically indexing and identifying textual data, using the thesaurus.

Description

PROCESS FOR THE AUTOMATED DEVELOPMENT OF A THESAURUS OF NAMED ENTITIES THAT MAY INCLUDE A PLURALITY OF HIERARCHICAL LEVELS, AND USE OF SUCH A THESAURUS.

The invention relates to a method for automatically developing from textual data a thesaurus of named entities, a named entity which may include a plurality of entities of different hierarchical levels, comprising: an extraction step, to starting from textual data, from a character string designating a named entity; a step of processing the character string to establish a list comprising at least one entity segment; a step of eliminating the list of entity segments already present in the thesaurus, to form a list of new segments; and a step of updating the thesaurus for each new segment of the list of new segments. The invention also relates to a method of automatic indexing and identification of textual data, using the thesaurus.

i

PROCESS FOR THE AUTOMATED DEVELOPMENT OF A THESAURUS OF NAMED ENTITIES THAT MAY INCLUDE A PLURALITY OF HIERARCHICAL LEVELS, AND USE OF SUCH A THESAURUS.

FIELD OF THE INVENTION

The present invention relates to automated processing of textual data. It more specifically aims at the exploitation of textual data for a fine analysis of the information that this data includes, in particular when this data relates to named entities which can have several hierarchical levels.

TECHNOLOGICAL BACKGROUND OF THE INVENTION

It is known to use automated means of extracting information from collections of textual data to form databases from which it is possible to conduct more in-depth analyzes. To constitute these bases, one can be led to locate in this original information categories of named entities (address, organization, range of products ...) and to recognize all occurrences of the same named entity.

The disparity of forms, graphic or syntagmatic, referring to the same named entity poses a problem. The origin of this well-known disparity: it can be a typographical or spelling error in the original information, problems related to the transcription and / or transliteration of certain terms, the disparity of sources of information and in particular the lexical heterogeneity of these sources. In an attempt to overcome this variability in form, there is generally provided, during the constitution of a database, for a standardization step in order to reduce all the names relating to the same entity to a standard form, called form. normalized.

In this regard, we can refer to the document "normalization of named entities: combining declarative rules, endogenous resources and user-centered process" by V. Andréani et al, Canadian Journal of Information Science and Library Science, Volume 35 (3) (pp. 229-263).

Certain named entities can have several hierarchical levels. This is for example the case of organizations such as companies (head office, subsidiary, management, services); educational and research establishments (university, laboratory, service); or even addresses of places (country, city, street).

It is generally not known in advance how text data will be used when it has been organized into a database. In particular, and concerning the named entities that can have several hierarchical levels, we cannot predict in advance according to which hierarchical level we wish to select the data from the database.

Thus, the aforementioned prior art document proposes to take the highest hierarchical level as the standardized form of the named entity. For example, and according to this document, the entity named "Laboratory Y of University X" is standardized in "University X". This choice does not allow a detailed analysis to be carried out from the database, for example to search for all the information relating to Laboratory Y of University X.

Alternatively, one could choose to take the lowest hierarchical level as the standard form. We could for example normalize the named entity of the previous example into "Laboratory Y, University X". However, this approach does not make it easy to conduct an analysis on a higher level (search for all the information relating to University X) except by manually grouping the information extracted for each of the named entities which is linked to the same entity. higher level. However, this work can be laborious and must be repeated for each new request.

OBJECT OF THE INVENTION

The present invention aims to overcome all or part of the aforementioned drawbacks. It aims in particular to allow the selection of records from a database according to different hierarchical levels of a named entity, without requiring a reprocessing of the database or the manual grouping of standardized forms of entities attached hierarchically to a same entity. In other words, the invention aims to provide a tool allowing a user to choose, at will, the hierarchical level of work in which he wishes to evolve.

BRIEF DESCRIPTION OF THE INVENTION

With a view to achieving one of these aims, the object of the invention provides a method of automatic elaboration from textual data of a thesaurus of named entities, a named entity being able to comprise a plurality of entities of different hierarchical levels, the thesaurus forming a database made up of a plurality of entity records, each record associating for a determined entity:

- a standardized form of the specified entity;

- at least one character string extracted from the textual data and designating the determined entity;

- and, if applicable, at least one link to another record in the thesaurus corresponding to an entity with which the determined entity is linked hierarchically.

According to the invention, the process for the automatic production of the thesaurus comprises:

- a step of extracting, from the textual data, a character string designating a named entity;

a step of processing the character string to establish a list comprising at least one entity segment;

- a step of eliminating the list of entity segments already present in the thesaurus, to form a list of new segments;

- a step of updating the thesaurus for each new segment of the list of new segments

According to other advantageous and non-limiting characteristics of the invention, taken alone or in any technically feasible combination:

• the processing step comprises a step of delimiting at least one entity segment by identifying punctuation marks in the character string;

• the processing step includes a step of determining the hierarchical level associated with each entity segment;

• the processing step includes a step of rewriting each entity segment;

• the processing step includes the identification of standard primers in each entity segment, the standard primers being contained in a processing lexicon;

• the processing lexicon associates a standard primer with a hierarchical level and a rule for rewriting the standard primer in an entity segment;

• the standard leader is an empty word and the rewrite rule consists in deleting the empty word in the entity segment;

• the method comprises a step of correcting errors of the entity segments;

• the step of updating the thesaurus includes a step of creating a new record;

• the step of updating the thesaurus includes a step of associating a new segment with a preexisting record;

• the update step is subject to validation and / or prior modification by a user;

According to another aspect, the invention relates to a method for automatic indexing of textual data comprising:

a step of extracting from the textual data a character string designating a named entity which may include a plurality of entities of different hierarchical levels;

a step of processing the character string to establish a list comprising at least one entity segment;

- a step of searching, in a thesaurus of named entities established using the method of elaborating the invention, of a matching record for each entity segment of the list;

a step of linking, in an indexing table, the matching record or a reference to the record matching with the text data or a reference to the text data.

The invention also relates to a method for automatic identification of textual data using an indexing table obtained according to the preceding method, in which the identification method comprises:

- a step of obtaining a designated entity to which the identification must relate;

- a search step, in the entity thesaurus, of the record corresponding to the designated entity;

a step of identifying, from the indexing table, the textual data associated with the record corresponding to the designated entity.

Finally, the invention relates to a computer program containing instructions adapted to the implementation of each of the steps of one of the methods presented above, when the program is executed on a computer.

BRIEF DESCRIPTION OF THE DRAWINGS

Other characteristics and advantages of the invention will emerge from the detailed description of the invention which will follow with reference to the appended figures in which:

Figure 1 shows an example of a thesaurus developed according to a method according to the invention;

FIG. 2 represents a sequence of steps of an automatic indexing method of textual data according to the invention;

FIG. 3 represents a sequence of steps of a method for automatic identification of textual data in accordance with the invention;

FIG. 4 represents a sequence of steps of a process for the automatic production of a thesaurus according to the invention.

DETAILED DESCRIPTION OF THE INVENTION

Thesaurus of entities.

The present invention provides for developing a thesaurus of entities, each named entity being able to include a plurality of entities of different hierarchical levels. As will be detailed in detail later, this thesaurus is produced automatically from a collection of source textual data.

"Textual data" means any information represented in the form of text. It can thus be scientific and technical information, as contained in patents, scientific articles. This information can be unstructured or structured (for example organized in a database, in which predetermined fields at least partially organize the information).

By "entity" is meant a hierarchical level of a named entity having several hierarchical levels. Thus the entity named "Laboratory X of University Y" includes the two entities "Laboratory X" and "University Y". A named entity that has only one hierarchical level, for example "Company Z", has only one entity, "Company Z". To keep all its generality in the present description, it will be considered that the named entities necessarily have a plurality of hierarchical levels. However, the invention is in no way limited to this particular type of named entity, and the treatments proposed apply to any type of named entity, whether or not it comprises a plurality of hierarchical levels.

A thesaurus according to the invention takes the form of a database made up of a plurality of records, each record describing a specific entity. Such a record includes at least the following fields:

- a designation field to receive a standardized form from the specified entity.

- a text field to receive at least one character string extracted from textual data, and designating the determined entity.

- and, if necessary, a link field to receive a link to another record in the thesaurus, this other record corresponding to an entity with which the entity determined and hierarchically linked.

For each record in the thesaurus, the designation and text fields are necessarily completed. The link field is filled in as far as the corresponding entity is hierarchically linked to at least one other entity, also recorded in the database.

Each entity can be linked to one or more other entities. Consequently, the link field can include one or more links, each link pointing to another record in the thesaurus. The link can be interpreted as the relation "belongs to". More generally, the presence of a link between two records makes it possible to determine the nature of the hierarchical relationship existing between two entities, that is to say which of the two entities is hierarchically superior to the other.

The designation field can be defined as a character string. The text field can be defined as at least one character string. This string, extracted from textual data, may have received processing, that is to say that it is not necessarily plain text directly extracted from source textual data.

ίο

The records forming the database may also provide additional fields. They can thus include an index field allowing to easily designate and computerize the records of the database.

As we ask, it appears under textual sources evoked in the introduction of this indeed possible that the same entity different forms in the data By attaching these different forms in the text fields of the same record in the database constituting the thesaurus, an exhaustive and compact database is formed.

By way of illustration, the original textual data is a collection of scientific articles. The named entities correspond to the name of the organization with which the authors of these articles are affiliated. One of the documents in the collection shows the organizational name "Glass Surface and Interfaces, Joint Research Unit (UMR) 125, National Center for Scientific Research / Saint-Gobain, Aubervilliers, France". After the automatic processing which will be detailed later in this description, there is a thesaurus integrating this data and an extract of which is represented in FIG. 1.

A first record, of index 1, includes the designation “Surface Glass Interface” as a standardized form of the entity and is linked to the index records 2 and 3 which will be described. In the example of FIG. 1, the thesaurus also presents a text field, which for this index record 1 is filled in by the processed form (by deleting hollow words) from the initial character string "Surface Glass Interfaces, Joint Research Unit ”. The next record, index 2, includes the designation "CNRS" as the standard form of the entity. The existence of a link pointing to this record makes it possible to establish that the entity “Surface Glass Interface” is of a level lower than the entity “CNRS”. The plain text field of index 2, includes several expressions

The recording (which could have been extracted from textual data during the indexing of several scientific articles) "CNRS", "C.N.R.S", "National Center for Scientific Research".

The record of index 3 is similar to that of index 2, and for the sake of brevity, we will not detail its content.

Automatic indexing of source text data

The invention takes advantage of this data structure to propose an automatic indexing process for textual data.

The method aims to establish an index of the textual data processed, also called "indexing table", that is to say a data structure associating a portion of the textual data (a source document, an entry of a database source data, a tag in unstructured textual data) to the entities that this portion contains. The indexing table allows you to quickly identify the portion (s) of textual data that includes a specific named entity. Of course, the indexing table may have other entries than that corresponding to the portions of textual data and to the entities, but the present description will be limited to these two fields for the sake of simplification.

With reference to FIG. 2, a method of automatic indexing of textual data in accordance with the invention comprises a first step of extracting, from a portion of the source textual data, a character string designating a named entity which may include several hierarchical levels. In Figure 2, this character string is "XYZ, ABC" and represents a named entity consisting of the two entities "XYZ", and "ABC".

When the source textual data is structured, for example in the form of a database or of markers in the text, the extraction step therefore consists in querying the base or locating the marker in the textual data in order to obtain in return the character string. In the case where the source textual data is not structured, we can rely on known methods of automatic recognition of named entities, for example that associating the use of a formal grammar, statistical models, and / or of entity category databases.

The automatic indexing process continues with a step of processing the character string to identify at least one entity, that is to say at least one hierarchical level of the named entity. This processing includes segmenting the character string into one or a plurality of segments. Each segment, called entity segment in the rest of this presentation, corresponds to an entity, that is to say at a hierarchical level of the named entity. The step of processing the character string leads to establishing a list comprising at least one entity segment. In Figure 2, this list is ["XYZ", "ABC"], consisting of the two segments "XYZ" and "ABC".

This processing can include a step of delimiting at least one entity segment by identifying punctuation marks in the character string. In other words, it involves cutting the character string extracted in the previous step into one or a plurality of segments, by assuming that the different hierarchical levels of the named entity are separated by such signs (commas, period, etc.). The delimitation can be based on another criterion, such as the detection of a particular pattern or leader in the character string.

The processing can also include a step of rewriting the character string and / or the segments to replace certain terms encountered in a standardized form, or to simply eliminate them in particular when they are hollow words, such as prepositions or articles.

For example, when the named entities are from the category of addresses, the character string "4 rue Béridot, 38500 Voiron, France" could be processed during the processing step to provide the list of following segments [ 4; R. Béridot; 38500; VOIRON; FR] . In addition to the reorganization of the segments, we note that the terms "France" and "Rue" have been rewritten in the standardized forms "FR" and "R.". We also note that the processing was based on the punctuation marks and the detection of determined primers (of the type "C + Street L +" or "CCCCC", where C represents a digit, C + a sequence of digits, and L + a sequence of letters) to correctly delimit the street number and the postal code.

In general, the automatic processing step can be based on a processing lexicon, associating standard primers with delimitation and / or rewriting rules. To use the previous example, the processing lexicon makes it possible to identify the primers types "France" and "Street" of the character string and to rewrite them in their standardized forms "FR" and "R.". The processing lexicon also makes it possible to identify a sequence of 5 digits in an address as forming a postal code, this sequence forming an entity segment, and the characters following this sequence as another entity segment designating the city. The treatment lexicon can also include a list of hollow words. The rewrite rule associated with these hollow words then consists in eliminating them from the entity segment. To be complete, and although this characteristic is not necessarily exploited during the execution of the indexing process, the processing lexicon can associate each standard primer to a hierarchical level (for example associate the term "France", to its form rewritten "FR" and at the hierarchical level "1"), in order to be able to establish the hierarchical relationships existing between the different identified segments. Exploitation of the processing lexicon therefore also makes it possible to identify the hierarchical relationship that may exist between the entity segments.

Whatever the treatments actually carried out on the character string designating the named entity, there is available at the end of the processing step a list comprising at least one entity segment.

With reference to FIG. 2, the automatic indexing process continues with a search step, in the text field of the entity thesaurus, of each entity segment of the list of segments so as to identify, for each segment, a corresponding record. In FIG. 2, this step returns the recording indexes 1 and N, for which the respective segments “XYZ” and “ABC” are actually found in the text field.

In favorable situations, which should be most numerous once the thesaurus is sufficiently constituted, this step leads to providing for each segment of the list, the thesaurus record or a reference to a record (a pointer to this record, a number of index of this record) which corresponds to it.

Unfavorable situations occur when the search in the thesaurus does not identify a record corresponding to one of the entity segments in the list. This situation can have several origins.

Thus, the segment can present an error, for example typographic or orthographic. In order to remedy this situation, the method may include an error step. The on the chain error correction of raw characters, processing, but preferably can before it correction apply the step of applies on the entity segments of the list of segments.

There are many possible error correction methods, but advantageously the error correction step includes a measure of similarity between an entity segment and the strings of the text field of the thesaurus records. If, for a particular thesaurus record, this similarity measure is less than a determined threshold, the entity segment is considered to have an error, and it is replaced by the character string of the text field of the particular record .

can be in this as a executed before case executed step from sub step

The correction stage the research stage and systematically, for example the treatment stage. Alternatively, this correction can be executed after the search step, only in the case where this search step was not successful for an entity segment, that is to say that the search for the entity segment in the thesaurus produced no matching records.

In some cases, it is also possible that the thesaurus does not contain an entity from the list, even after having performed an error correction step on the entity segment (s). In this case, you can choose to update the thesaurus for example by adding a new record corresponding to this entity or by attaching a segment to an already existing record in the thesaurus. This situation of updating the thesaurus, represented in dotted lines in FIG. 2, will be the subject of a detailed description in the part of the presentation relating to the constitution of the thesaurus. In a variant, also detailed below, the update of the thesaurus is validated, edited or carried out by a user.

are

Whatever the chained, there are particular stages which are at the end of the stage of searching for references to thesaurus records corresponding to the entity segments of the list, ie to each entity identified in the chain of characters.

In a following step, an indexing table is formed, that is to say that the portion of the source data (or a reference to this portion) being indexed is linked to the records identified in the thesaurus to the outcome of the previous step. Thus, and as shown in FIG. 2, at the end of this linking step, there is an indexing table which can have two columns, the first column making it possible to designate the portion of the source textual data. which has just been indexed (document, entry of a database, etc.), and a second column allowing to designate the thesaurus records identified during the process steps which have just been presented. Each row of the indexing table therefore associates a portion of the text data and at least one reference to at least one entry from the entity thesaurus.

When there is, as is generally the case, a plurality of portions of textual data to be indexed, the automatic indexing method which has just been presented for each of these portions can be repeated, so as to form an indexing table connecting each portion of textual data to the different hierarchical levels of entities that this portion contains.

Automatic identification

According to the invention, the indexing table can be exploited in a method of automatic identification of a portion of textual data, among the textual data having served as a basis for constituting the index.

This identification method, represented in FIG. 3, comprises a first step of obtaining a designated entity to which the identification must relate. This may, for example, involve obtaining from a user the designation of the entity serving as the basis for the section of data they wish to operate. In Figure 3, the entity "ABC" was provided during this step.

The designated entity can have any hierarchical level: to continue one of the previous examples, the user can search, when the source textual data consists of scientific articles, articles for which at least one of the authors is affiliated to the CNRS (higher hierarchical level) or search for articles for which at least one of the authors participates in a particular laboratory of the CNRS (lower hierarchical level)

The identification process continues with the search step in the text field of the entity thesaurus, of the record corresponding to the designated entity. In Figure 3, this step provides a reference to the N index record of the thesaurus.

Using the index of this record, one can find, during an identification step, in the indexing table all the portions of textual data associated with this record. In FIG. 3, this step provides a reference to the document D1, to indicate that this document refers to the entity designated "ABC".

Thus by combining the exploitation of the entity thesaurus and the indexing table, it is possible to carry out a search in textual data associated with a named entity which can have several hierarchical levels, by making the search relate to n ' no matter what hierarchical level of the named entity.

Automatic thesaurus development

As we have seen, the thesaurus forms a database made up of a plurality of entity records. Each record combines for a specific entity:

- the standardized description of the entity determined in a designation field;

- at least one character string, extracted from textual data, and designating the named entity, in a text field;

- and, where appropriate, at least one link to another record in the thesaurus corresponding to an entity with which the determined entity is hierarchically linked. This link or these links are placed in the link field.

The invention proposes a method for automatically developing the thesaurus of named entities, a named entity being able to comprise a plurality of entities having different hierarchical levels. The steps of this process are schematically represented in FIG. 4.

The method successively processes portions of source textual data, for example documents from a collection of documents. For each portion, the method comprises a step of extracting, from the source textual data portion, a character string designating a named entity. The automatic production method also includes a step of processing the character string to establish a list comprising at least one entity segment.

These extraction and processing steps are identical in all respects to that described in the context of the detailed description of the automatic indexing method, and represented in FIG. 2. For the sake of brevity their description is not repeated here .

To establish a compact and coherent thesaurus, and as has been explained above, the processing step advantageously comprises a step of rewriting and / or correcting the error of the segments of entities.

At the end of these steps, we therefore have a list of segments, some of which may already be present in the thesaurus. The process for automatically developing the thesaurus therefore includes a step of eliminating this list of entity segments, those already present in the thesaurus. More specifically, and for each segment of the list of entity segments, we look for their presence in the text field of the thesaurus. If there is a corresponding record in the thesaurus, the segment is not selected from the list of new segments. A refined list is thus formed, called "list of new segments". In FIG. 4, this list is made up of the new segment “ABC”.

For each new segment of this list, the thesaurus is updated.

It can be the creation of a new record in the thesaurus, during a creation step. The creation of a record consists in filling in the database forming the thesaurus the designation and text fields (using the entity segment identified) and, if necessary, the link field.

Alternatively, it may be a question, during an association step, of associating a new segment with a preexisting recording. Adding the corresponding new character string to the text field of a record, the segment error correction step may reveal that a new segment is not similar enough to consists of a Par segment. example, of the existing record field to be able to be corrected in this form, but that however it presents enough similarity so that, in any hypothesis, the new segment the same entity as that corresponding to this It is preferable in this case to proceed to new to this record, to the creation of a new text character string of a refers to record the association of the segment rather than to proceed record. Associating a new segment with an existing record consists in adding the character string corresponding to the new entity segment to the text field of the record. In FIG. 4, this association step has led to the updating of the text field of the record N, with the string "ABC".

It cannot be guaranteed that the fully automatic processing of textual data will systematically lead to a perfectly coherent thesaurus. Thus, it is possible that certain typographical errors cannot be corrected purely automatically. In the case where it is important to have a very coherent thesaurus, it can be provided that the updating of a new thesaurus record is subject to prior validation, and possible editing, by a user.

Thus, during the creation step, a user may prefer to provide a standardized form of the entity (designation field) different from that proposed (which corresponds to the identified entity segment). For example, a user might prefer to use the character string "CNRS" as the content of the designation field rather than the string "C.N.R.S. Which would form a new segment.

In a second example, and more substantially, a user might prefer to associate a new segment with an existing record rather than creating a new record. For example, a user might prefer to associate a new segment "ST Gobain" with a preexisting record whose designation and text fields would be "Saint Gobain".

According to a last example, a user could wish to confirm an association of a new segment with an existing record which the method of the invention automatically proposes. In this case, he could validate the proposal or, on the contrary, choose to associate the new segment with another record in the thesaurus.

The processes of preparation, indexing and identification which have just been presented are called upon to be implemented by a computer. The invention therefore also relates to the computer program or programs containing instructions adapted to the implementation of each of the steps of the abovementioned methods when the program is executed on a computer.

The invention finds a particularly interesting application in the field of processing scientific and technical information (scientific articles, conference summaries, conferences, patents, ...); but can also be applied to the processing of information from any application domain, in particular if this information is available in the form of textual data, and that it is capable of incorporating named entities having several hierarchical levels.

Of course, the invention is not limited to the example of implementation described and it is possible to make variant embodiments without departing from the scope of the invention as defined by the claims.

Thus, although it has been indicated that certain data could be organized in the form of a table, the invention is certainly not limited to such a data structure, and it can be implemented independently of the data structure chosen. We can thus choose to structure the data in the form of lists, indexed lists, tables, objects, databases or any form of structure, depending on the need and the circumstance.

Claims (13)

1. Method for the automatic production from textual data of a thesaurus of named entities, a named entity possibly comprising a plurality of entities of different hierarchical levels, the thesaurus forming a database made up of a plurality of entity records, each record combining for a specific entity: - a standardized form of the specified entity; - at least one character string extracted from textual data and designating the determined entity r - and, where applicable, at least one link to another record in the thesaurus corresponding to an entity with which the determined entity is hierarchically linked; the process of automatic development of the thesaurus comprising: - a step of extracting, from the textual data, a character string designating a named entity; a step of processing the character string to establish a list comprising at least one entity segment; - a step of eliminating the list of entity segments already present in the thesaurus, to form a list of new segments; - a step of updating the thesaurus for each new segment of the list of new segments. 2. Method according to the preceding claim wherein the processing step comprises a step of delimiting at least one entity segment by identifying punctuation marks in the character string. 3. Method according to one of the preceding claims wherein the processing step comprises a step of determining the hierarchical level associated with each entity segment. 4. Method according to one of the preceding claims wherein the processing step comprises a step of rewriting each entity segment. 5. Method according to one of the preceding claims, in which the processing step comprises the identification of standard primers in each entity segment, the standard primers being contained in a processing lexicon. 6. Method according to the preceding claim, in which the processing lexicon associates a standard primer with a hierarchical level and with a rule for rewriting the standard primer in an entity segment. 7. Method according to the preceding claim in which the standard primer is an empty word and the rewrite rule. is entity. to delete the empty word in the segment 8. Process according to one of the claims previous including a step of correction of errors of segments entity. 9. Method according to one of the preceding claims wherein the step of updating the thesaurus comprises a step of creating a new record. 10. Method according to one of claims 1 to 8 wherein the step of updating the thesaurus comprises a step of associating a new segment with a preexisting record. 11. Method according to one of the preceding claims, in which the updating step is subject to the validation and / or to the prior modification of a user. 12. Method for automatic indexing of textual data comprising: a step of extracting from the textual data a character string designating a named entity which may include a plurality of entities of different hierarchical levels; a step of processing the character string to establish a list comprising at least one entity segment; a step of searching, in a thesaurus of named entities established using a method in accordance with one of the preceding claims, of a matching record for each entity segment of the list; a step of linking, in an indexing table, the matching record or a reference to the record matching with the text data or a reference to the text data. 13. Method for automatic identification of textual data using an indexing table obtained according to a method according to the preceding claim, in which the identification method comprises: - a step of obtaining a designated entity to which the identification must relate; - a search step, in the entity thesaurus, of the record corresponding to the designated entity; - an identification step, from the table 5 indexing, text data associated with the record corresponding to the designated entity. 14. Computer program containing instructions adapted to the implementation of each step of a 10 Method according to one of claims 1 to 13, when the program is executed on a computer.