FR2876198A1 - Thematic summary generating method for computing device, involves selecting set of words of document with respect to words of document having largest weight so as to automatically generate thematic summary based on selected set - Google Patents

Abstract

The method involves weighing each word found from a document e.g. digital file, by a weight determined with respect to one of units among a number of repetitions of the word in the document and a respective thematic factor. A set of words of the document is selected with respect to the words of the document having the largest weight so as to automatically generate a thematic summary based on the selected set of words. Independent claims are also included for the following: (A) a computing device for automatically generating a thematic summary (B) a computer program on an information medium for automatically generating a thematic summary.

Description

Automatic generation of thematic summary

  The present invention relates to a method for automatic generation of thematic summary of document at least textual.

  In the state of the art, systems for automatic generation of summary of a document rely in particular on linguistic and statistical analyzes of the document to be summarized. These systems constantly generate the same summary for a given unmodified document. Users can not interact with the summary generation, for example to direct the summary to a more specific theme of the document.

  The invention aims to overcome the aforementioned drawbacks by generating different summaries of the same document to automatically obtain thematic summaries of the document.

  To achieve this objective, a method for automatically generating a thematic summary from an at least textual document in a computing device, comprising a step of determining at least one thematic parameter of the document, is characterized in that it comprises the steps following of.

  selecting at least one thematic parameter from specific thematic parameters, finding words in the document corresponding to each selected thematic parameter in order to associate a respective thematic factor with each found word corresponding to a selected thematic parameter, weighting each word with less found of the document by a weight determined according to at least one of a number of repetition of the word in the document and at least one respective thematic factor, and - selecting a set of word of the document according to the words of the document with the highest weights to generate the thematic summary depending on the selected word set.

  In the step of weighting, the theme factor of each found word corresponding to a given selected thematic parameter is increased or decreased, respectively to a maximum or minimum value, of thematic factor so that the summary generated is close to, or distant from, the parameter selected theme given.

  The invention also relates to a computing device for automatically generating a thematic summary from an at least textual document comprising means for determining at least one thematic parameter of the document. The device is characterized in that it comprises: a means for selecting at least one thematic parameter among specific thematic parameters; a means for finding words in the document corresponding to each selected thematic parameter in order to associate a thematic factor; respective to each found word corresponding to a selected thematic parameter; means for weighting each at least one word of the document by a weight determined according to at least one of a number of repetition of the word in the document and minus a respective thematic factor; and - means for selecting a word set of the document based on the words of the document having the highest weights in order to generate the thematic summary dependent on the selected word set.

  Other features and advantages of the present invention will appear more clearly on reading the following description of several preferred embodiments of the invention, given by way of non-limiting examples, with reference to the corresponding appended drawings in which: FIG. 1 is a schematic block diagram of an automatic summary generation system implementing a method of automatic generation of thematic summary of a document according to the invention; and FIG. 2 is an algorithm of the method for automatically generating a thematic summary of a document according to the invention.

  In the rest of the description, a document is

  a digital file containing at least text. It may possibly comprise at least one image and / or at least one video sequence. A document is for example a page in HTML (HyperText Markup Language) or a document from a word processor.

  A thematic summary of a document according to the invention is a summary oriented towards a theme or remote from a theme. A summary is a group of sentences that are relatively small in size relative to the document. A theme is defined by a set of words in a particular syntactic context, said set of words comprising a word or several words within the meaning of the invention.

  A thematic parameter according to the invention is a parameter influencing the generation of a thematic summary. A thematic parameter is often linked to a presence or absence character of it.

  For example, a thematic summary generated in dependence on the presence of the thematic parameter "sport" will tend to include words of the document relating to "sport", and conversely, a thematic summary generated in dependence on the absence of the thematic parameter "sport". "will tend not to include a word relating to" sport "and is thus removed from the thematic parameter. In another example, if a thematic parameter is the absence of the "temporal" theme, the generated summary will tend not to include any temporal data, such as date, time and year; or if a thematic parameter is the presence of the "number" theme, the generated summary will tend to contain numbers of the document.

  For example, the method is implemented for a document search service on the internet, presenting the search result in the form of a list of thematic summaries of documents according to user-selected thematic parameters. As a variant, the summaries provided in the list are general and independent of a theme, and then the search service proposes to the user to select a list of thematic parameters and names extracted from the documents found, and dynamically generates thematic summaries. customized according to the selected thematic parameters, as will be seen below.

  With reference to FIG. 1, the automatic summary summary generation system mainly comprises a thematic summary generation server SR, an SBD document database server, a SBP thematic database server, a database server. SB summary database and at least one user terminal T. The summary summary generation server SR mainly comprises a central processing unit UC, a document parser PD, a linguistic parser AL, a named entities extractor EE, a MT theme determination module and MR summary module.

  The SBL linguistic database server includes data such as thesauri that map thematic parameters to sets of words, particular lexicons, and so on. The user terminal T is connected to a respective access network RA by an LT link. It necessarily comprises display means and / or means for sound reproduction of digital data. The terminal T is for example a mobile radio terminal Tl, the link LT is a radiocommunication channel, and the respective access network RA comprises the fixed network of a cellular radio network, for example of the GSM type (Global System for Mobile communications) with a General Packet Radio Service (GPRS) or UMTS (Universal Mobile Telecommunications System).

  According to another example, the terminal T is a personal computer T2, connected directly by modem to the line LT xDSL line or ISDN (Integrated Services Digital Network) connected to the corresponding RA access network.

  In another example, the terminal T is a fixed telecommunication terminal T3, the link LT is a telephone line and the respective access network RA comprises the switched telephone network.

  According to other examples, the user terminal T comprises a device or electronic personal telecommunications object to the user which can be a PDA communicating personal digital assistant. The terminal T may be any other portable or non-portable home terminal such as a video game console, or a smart TV receiver cooperating with a remote control display or an alphanumeric keyboard also used as a mouse through an infrared link.

  In another example, the access network RA comprises a connection network of several user terminals.

  The user terminals T and the access networks RA are not limited to the above examples and may be constituted by other known terminals and access networks.

  The SBD, SBP and SBL database servers communicate with the thematic summary generation server SR through an RT telecommunications network, such as the Internet, connected to the RA access networks.

  In a variant, at least one of the SBD, SBP and SBL database servers communicates locally with the thematic summary generation server SR.

  In other variants, the data in the SBD, SBP and SBL database servers are distributed in one or two database servers.

  With reference to FIG. 2, the method for automatically generating a thematic summary of a document originally stored in the document database server SBD comprises, according to the invention, steps E1 to E7 automatically executed in the server. SR topic summary generation.

  In step E1, the document parser PD transforms the document into a pre-parsed document, so that other modules in the thematic summary generation server SR use and interpret the pre-parsed document regardless of the format of the document. document. Pre-analysis involves analyzing the document and creating from the document and analysis a pre-analyzed document containing and describing the various elements of the document. An element of the document is for example a paragraph, a title, an image, a table, or a word. The pre-parsed document contains descriptions of items such as word underlining, word fat writing, image locations, video clips, bullets, and so on. The substeps E21 to E23 consist in determining thematic parameters of the document. Some of these steps can be modified or supplemented or deleted, and other steps can be added to determine thematic parameters.

  In step E21, the linguistic analyzer AL analyzes the pre-analyzed document to determine lexical characteristics of the type of word lemmas used and topics addressed in the document, of the syntax characteristics of the type of grammatical functions of words used and division of sentences into nominal groups. and verbal, and semantic features. These characteristics are stored in correspondence with the document in the SBP thematic parameter database server to determine at least one thematic parameter of the document in the following steps E22 and E23.

  In step E22, the named entity extractor EE extracts named entities, i.e. first and last names of persons, place names, trademarks and company names, from the document analyzed according to the lexical, syntactic and semantic characteristics determined and provided by the linguistic analyzer AL. The named entities are stored in correspondence with the document in the SBP Thematic Parameter Database Server.

  In a variant, the method according to the invention is not limited to the named entity extraction method described above, but accepts all known named entity extraction methods.

  In step E23, the subject determination module MT determines a majority theme or themes addressed by the pre-analyzed document based on statistical measurements made on the lexical, syntactic and semantic characteristics and possibly on the basis of the information contained in a document. thesaurus. The theme is stored in correspondence with the document in the SBP Thematic Parameter Database Server.

  In a variant, the method according to the invention is not limited to the theme determination method described above, but accepts all the known topic determination methods.

  Named entities and themes are examples of thematic settings in the document.

  In a variant, the step E1 is deleted and the thematic parameters are directly determined from the document. In another variant, the steps E1 to E23 which take place only once, are preceded by a step of selecting the documents.

  In step E3, the central unit UC selects at least one thematic parameter from among the thematic parameters determined in steps E22 and E23. The selected thematic parameters are stored in the SBP thematic parameter database server in correspondence with the document.

  For example, during a web service for generating a summary of a document, the user terminal T selects, for a given document, thematic parameters presented to the terminal T by the summary summary generation server SR in the form of a document. an HTML page containing a form. For a document dealing with tourism in Paris, the parameters "weather" and "Paris" are selected from the following words. Paris, restaurants, theaters, shows, Eiffel Tower, sun, Victor Hugo, sports, numbers, dates, weather.

  In the example of the search service, the thematic summary generation server SR derives thematic parameters selected from the keywords entered by the user of the terminal during his search. In another example similar to the last one, the user of the terminal enters a sentence in natural language, and a linguistic analyzer extracts key words from the sentence entered as thematic parameters.

  According to a combination of the preceding examples, the summary summary generation server SR provides summaries of the found, general and independent documents of a theme, and then proposes to the user for each document found a form so that thematic parameters extracted from this document are selected. The user then selects at least one parameter using the form that the server SR receives. The server generates summaries according to the parameters selected by the user, which summaries are displayed in the user terminal T. In step E4, the central unit UC searches in the pre-analyzed document for words corresponding to each one. Thematic parameters selected in the previous step E3. For this word search, the central processing unit UC consults the thesauri stored in the database server SBL and reads keyword lists previously associated with the selected thematic parameters. For example, a keyword list includes the keywords "city", "mode", "luxury" and "eiffel tower" when the thematic parameter is "Paris". However, a keyword list can include only one keyword, such as the thematic parameter when it consists of a single word. Then, the central unit UC compares each word of the document to the keywords of the list associated with each selected thematic parameter in order to associate a thematic factor corresponding to the selected thematic parameter, to each found word identical to a keyword of the list.

  The association of each found document word and the respective thematic factor is stored in the SBP server.

  When a word of the document is not identical to any keyword in the list associated with each thematic parameter, the thematic factor is a predetermined constant, for example equal to zero.

  In step E5, the summary module MR determines a weight for each at least one document word based on a repetition number of the word in the document and / or the respective thematic factor or factors associated with the word. When the found word is associated with several respective thematic factors, the weight of the found word is for example the average of the respective thematic factors. In another example, the weight of the document word is the product of the repetition number of the word in the document and the respective thematic factor (s).

  The summary module MR weighting each document word by the respective weight thus determined. Note that a document word that is not found in the previous step E4 may also be weighted by a weight equal to its repetition number in the document. According to another variant, the weight of each document word that has not been found in correspondence with a selected thematic parameter may be a constant, for example equal to a minimum value of thematic factor such as zero, when the weights do not depend on than thematic factors.

  To emphasize the thematic orientation of the summary to a given selected thematic parameter, the summary module MR increases the theme factor of each found word corresponding to the given parameter to a maximum value of thematic factor so that the generated summary is close to the given parameter. . By thus increasing the weight of the found words corresponding to the given parameter, these words are privileged in the generation of the summary and thus direct the summary so that the theme of the latter is close to the given selected thematic parameter.

  In a variant of step E5, instead of emphasizing the thematic orientation of the summary to a given selected thematic parameter, the summary module MR decreases the theme factor of each found word corresponding to the given parameter to a minimum value of thematic factor so that the generated summary is far from the given parameter. For example, the predetermined minimum value is equal to the least of the weights determined in step E5. For example, the second predetermined value is zero. By decreasing the weight of certain words, these words tend to be removed from the generation of the abstract and tend not to appear in the abstract whose theme is thus far from the selected thematic parameter.

  In step E6, the summary module MR selects a word set of the document based on the words of the document having the highest weights in order to generate a thematic summary depending on the set of selected words and preferably in order to display in step E7 this summary in the user terminal T. A set of words participating in the generation of the summary is possibly a sentence or a paragraph.

  For example, the summary module MR selects, as sets of words, sentences based on the word weights of the document. For this purpose, the MR module adds the weights of the words in each sentence to produce the weight of each sentence and selects the sentences with the highest weight to construct the summary.

  In another example, the summary module MR selects sentences according to the weight of the words of the document, determines the weight of a link between two sentences, determines the weight of each sentence, in particular according to the weights of the words of the sentence, weights associated with the links concerned by said sentence and the number of sentences to which said sentence is linked, and selects the sentences having the highest weight to construct the summary. The weight of a link between two sentences can be determined by words in common to both sentences.

  The steps E3 to E7 can be repeated so that the thematic summary displayed is specified according to thematic parameters selected progressively by the user, as indicated by the dotted line loop in FIG.

  The automatic generation of the thematic summary according to the invention can be supplemented by other known steps, such as the reduction of a sentence, the selection of structural elements such as a title, etc. The invention described herein relates to a method and a device for automatically generating a thematic summary of a document. According to a preferred implementation, the steps of the method are determined by instructions of a program for automatically generating a thematic summary of a document incorporated in a computing device such as the summary generation server SR. The program comprises program instructions which, when said program is loaded and executed in the computing device whose operation is then controlled by the execution of the program, perform the steps of the method according to the invention.

  Accordingly, the invention also applies to a computer program, including a computer program on or in an information carrier, adapted to implement the invention. This program can use any programming language and be in the form of source code, object code, or intermediate code between source code and object code such as in a partially compiled form, or in any other desirable form to implement a method according to the invention.

  The information carrier may be any device or entity capable of storing the program. For example, the medium may comprise storage means, such as a ROM, for example a CD ROM or a microelectronic circuit ROM, or a magnetic recording medium, for example a diskette (floppy disc) or a disk hard. On the other hand, the information medium may be a transmissible medium such as an electrical or optical signal, which may be conveyed via an electrical or optical cable, by radio or by other means. The program according to the invention can in particular be downloaded to an Internet type network.

  Alternatively, the information carrier may be an integrated circuit in which the program is incorporated, the circuit being adapted to execute or to be used in the execution of the method according to the invention.

Claims (8)

  1 - Method for automatically generating a thematic summary in a computing device (SR) from an at least textual document in the form of a digital file initially stored in a database server (SBD, SBP, SBL) communicating with the computing device, comprising a step of determining (E22, E23) and storing at least one thematic parameter of the document, characterized in that it comprises the following steps of: - selecting (E3) and storing at least one thematic parameter among specific thematic parameters, finding (E4) and memorizing words in the document corresponding to each selected thematic parameter in order to associate a respective thematic factor with each found word corresponding to a selected thematic parameter, - weighting (E5) each word at least found the document by a weight determined according to at least one of a number of repetition of the m ot in the document and at least one respective thematic factor, and - selecting (E6) a set of word of the document according to the words of the document having the highest weights in order to automatically generate the thematic summary depending on the set of words selected.   2 - Process according to claim 1, characterized in that in the step of weighting (E5), the theme factor of each found word corresponding to a given selected thematic parameter is increased to a maximum value of thematic factor for that the generated summary is close to the given selected thematic parameter.   3 - Process according to claim 1, characterized in that in the step of weighting (E5), the theme factor of each found word corresponding to a given selected thematic parameter is decreased to a minimum value of thematic factor for that the generated summary is far from the given selected thematic parameter.   4 - Process according to any one of claims 1 to 3, wherein when a found word is associated with several respective thematic factors, the weight of the found word is the average of the respective thematic factors.   A method according to any one of claims 1 to 4, wherein the step of determining at least one thematic parameter comprises extracting (E22) at least one named entity from the document as a thematic parameter.   6 - Process according to any one of claims 1 to 5, wherein the step of determining at least one thematic parameter comprises a determination (E23) of at least one theme of the document as a thematic parameter.   7 - Process according to any one of claims 1 to 6, comprising previously a transformation (El) of the document into a pre-analyzed document describing elements of the document, and a determination (E21) of lexical, syntactic and semantic characteristics of the document. pre-analyzed document so that at least one thematic parameter of the document is determined (E22, E23) according to said characteristics.   8 - Computer device (SR, T) for automatically generating a thematic summary from an at least textual document in the form of a digital file initially stored in a database server (SBD, SBP, SBL) communicating with the computing device, comprising means (AL, EE, MT) for determining and storing at least one thematic parameter of the document, characterized in that it comprises: - means (UC) for selecting and memorizing at least one thematic parameter among specific thematic parameters; - means (UC) for finding and memorizing words in the document corresponding to each selected thematic parameter in order to associate a respective thematic factor with each found word corresponding to a selected thematic parameter, means (MR) to weight each at least one word found in the document by a determined weight based on at least one of a number of repetition of the word in the document and at least one respective thematic factor, and - means (MR) for selecting a set of word of the document according to the words of the document having the highest weights in order to automatically generate the thematic summary depending on the selected word set.   9 - Computer program on an information medium for automatically generating a thematic summary of an at least textual document in the form of a digital file initially stored in a database server (SBD, SBP, SBL) , said program comprising program instructions which when said program is loaded and executed in a computing device (SR) perform steps of determining (E22, E23) and storing at least one thematic parameter of the document and are characterized in that they perform the following steps: select (E3) and store at least one thematic parameter among specific thematic parameters, - find (E4) and memorize words in the document corresponding to each selected thematic parameter in order to associate a respective thematic factor with each found word corresponding to a selected thematic parameter, - weighting (E5) each word at least found from the ocument by a weight determined according to at least one of a number of repetition of the word in the document and at least one respective thematic factor, and - selecting (E6) a word set of the document according to the words of the document with the highest weights to automatically generate the thematic summary depending on the selected word set.