FR3030812A1 - AUTOMATIC ANALYSIS OF THE LITERARY QUALITY OF A TEXT ACCORDING TO THE PROFILE OF THE READER - Google Patents

Abstract

The invention relates to an improvement in the method of analysis of the literary quality of a text, implemented by a computer program according to the patent application FR 14 63074. The improvement consists in taking into account the profile of the reader.

Description

TECHNICAL FIELD The present invention relates to a method for automatically analyzing the literary quality of a text according to the profile of the reader. By "coded literary quality of a text", one understands within the framework of the invention, the literary quality of a text which is intrinsic to it and which materializes by a discrete note of 1 to 10, by a continuous score in [0, 1], as a real score 1-00, +00 [or as "very good", "good", "average", etc. labels.

By "known literary quality of a learning text" is meant in the context of the invention, the rating given by a reading expert profile to estimate the quality of it. STATE OF THE ART In general, the purpose of automatic text categorization is to teach a computer machine to classify a text in the correct category based on its content. Categorization algorithms can solve various problems of categorization of texts. Regarding the analysis of the quality of a literary or scientific text, different approaches have already been made and different categorization algorithms implemented. There are thus several works that concern the quality of a literary text, but most are not relevant because they define the concept of quality in a sense that is their own and therefore not really independent of the factors that are chosen.

One can quote the patent US7200606 in which the notion of quality is considered in the sense of relevance vis-à-vis a user request. Thus, one of the relevant approaches is the so-called intrinsic approach according to which it is a question of using categorization algorithms to classify documents according to textual characteristics (indicators) which are intrinsic to the text: composition, elements of style, precision vocabulary in relation to a subject, construction of reasoning, spelling, etc.

The characteristics of sorting depend on very variable orthographic, lexical and stylistic approaches, among which the length of the words, regularity of the vocabulary, analysis of the cooccurrences, use of the punctuation, detection of grammatical and orthographical errors, facility of reading, lexical links with a theme or genre, etc.

These characteristics related to the text can be usefully supplemented by semantic methods made around the relationship between quality and respect for orthographic and typographic rules, grammar (measurement of quality on long n-grams), capitalization, density of the text. text (relationship between letters and spaces) or its entropy (at the level of words, or even at the level of characters).

Lexicometry, a method of quantitative analysis of texts, can prove to be a useful tool for measuring quality or non-quality, paradoxically. Whatever methods and categorization algorithms are used, the main difficulty lies in the choice of indicators and the algorithm, and in their combination to evaluate the literary quality of a text.

There is little literature that focuses on the literary quality of a text by intrinsic approach. We can first of all cite the publications [1] and [2] which describe an extraction of intrinsic indicators from a raw literary text and then a regression or a classification to reach the target value that one seeks to determine. . The choice of indicators remains relatively brief, which makes it difficult to refine the quality analysis with a very good precision. The publication [3] discloses a prediction of quality from a small number of newspaper articles ("Wall Street Journal"). The analysis according to this publication remains basic, since only a correlation is established between each indicator and a target value carried out on about thirty reference articles. The applicant filed on December 22, 2014 under No. 14 63074 a French patent application relating to a method for analyzing the literary quality of a text, which makes it possible to ensure a better accuracy of the analysis.

There is still a need not taken into account which is to improve the analysis of literary quality according to the reader's profile, that is to say according to his tastes. The object of the invention is to respond at least in part to this need.

DISCLOSURE OF THE INVENTION To this end, the subject of the invention is a method for analyzing the literary quality of a text according to the reader profile, implemented by a computer program, comprising the following steps: / receive a plurality of readers said learning readers; to receive a plurality of texts, called learning texts; b / extract for each learning reader and for the learning texts, its own indicators that are the number of texts read, the ratio between its expression of interest and the number of texts read, and the averages of the notes given to the read texts; c / generate a vector representation of each learning reader from its own indicators; d / subjecting the different vector representations of the learning readers to a data partitioning classifier so as to obtain groups of learning readers; e / generate a vector representation for each group of learning readers by averaging the vector representations of the group's readers; b / extract the numerical indicators of each learning text, c / generate a vector representation of each learning text from its numerical indicators, f / for each group of readers, to learn between the components of the vector representation of the learning texts and vector representations of the groups of readers and the known literary quality of each text of learning, so as to obtain a predictive model of the literary quality according to the groups of readers; g / receive a new text to analyze; h / apply to the new text and from a given group of readers the predictive model built in step d /, so as to obtain the literary quality of the new text according to the given group of readers.

It is specified that according to the invention, the data partitioning ("data clustering" in English) makes it possible to obtain groups of readers, which therefore each define a profile of readers. The inventor of the present invention started from the observation that the invention according to the aforementioned patent application FR 14 63074 makes it possible to improve the accuracy of the literary analysis of a text, but that it was not completely satisfactory because it does not allow to have a quality of text according to the profile of the reader, that is to say according to his tastes. Also, the invention essentially consists in taking into account the reader's profile in the prediction of the quality of a literary text. The invention also relates to a computer program for implementing the method described above. DETAILED DESCRIPTION Other advantages and characteristics of the invention will emerge more clearly from a reading of the detailed description of exemplary embodiments of the invention, given by way of nonlimiting illustration with reference to the following figures: FIG. 1: representation schematic of the vector representation of a group of learning readers according to their profile; FIG. 2 is a flowchart of the learning steps of the method according to the invention implemented by computer program; FIG. 3 is a flow diagram of the steps for predicting the analysis of the subjective literary quality of the method according to the invention, also implemented by computer program. Subsequently, the terms "algorithm" and "computer program" which is the computer-readable coding of the algorithm are used interchangeably. Thus, an algorithm is an execution plan for a computer. The computer takes incoming data, applies the processing described by the algorithm and returns a result to the user. In the context of the invention, the algorithm implemented for the predictive analysis is a machine learning algorithm. In this type of algorithm, his decision rules are not fixed to the design, because it is designed so that he can modify his decision rules, according to the data he sees. The process itself according to the invention comprises three successive phases, the first being a "data clustering" phase, the second being a learning phase and the last being a successive prediction phase. First of all, we realize the clustering phase. We first consider all the metadata included in each learning text that are for example: - the literary genre: romance, drama, policeman, haiku, alexandrine ... - the age of the reader of the text: 6, 8 , 10 ... years, - emotions expressed, that is to say, who are felt by the characters of the texts. With regard to the emotions considered, the following six are preferably chosen: happiness, affection, interest, sadness, melancholy, anger and fear.

For a given reader, one then takes into account measures of his taste, that is to say indicators that are specific to him, for each metadata of each text of learning. These indicators, which are specific to it, are established as follows: - the number of these texts read; - the ratio between its expression of interest ("like" in English) and the number of texts read averages, - the average of its notes. The algorithm then generates a vector representation of each learning reader from its own indicators. The learning readers are then grouped into n groups according to a usual "data clustering" technique, by proximity of the vector representations. By way of example, n is equal to 20. It is specified that proximity is to be considered in the mathematical sense of the term, that is to say that it is obtained by calculating the cosine between vector representations. For each group of readers which thus determines a profile of readers, a vector representation for each group of readers is generated by averaging the vector representations of the readers of the group (step Si).

In parallel, the algorithm then performs the following steps, from the learning texts to be analyzed. The numerical indicators are extracted from the learning texts (step S'O). Then, a vector representation of each learning text is generated from its digital indicators (step S'1). The algorithm can advantageously proceed in the following manner for the construction of the vector representation, from a raw text to be analyzed. It generates several vector sub-representations of the received text to obtain low-level indicators.

The first under-representation consists of a word bag representation that analyzes the distributions of each word and analyzes the distributions of certain unigrams, bi-grams, 3-grams, 4-grams, 5-grams, and 6-grams. at the scale of the word and characters. Thus, in this step, the text is transformed into a series of tokens according to regular expressions of cutting. Word-bag representation does not take into account the formatting of text, the order of words, their meaning or relationships structured by words of connection. The second under-representation represents the morphosyntactic structure, which calculates the parameters of grammatical word distributions in the text and analyzes the distributions of each syntactic function in text, paragraphs, sentences, and propositions. Grammatical words are articles, prepositions, non-qualifying adjectives. The calculation of the parameters of the distribution of the grammatical words is made from criteria chosen from the mean, the variance, the standard deviation, the entropy, the distance between the distributions or a combination of these. A syntactic function is a verb, a noun, an adjective, an adverb, a determinant, a preposition. Thus, this step makes it possible to extract elements of structure of the text in so far as to go up to the pragmatic level of the general comprehension of the text. The third underrepresentation represents writing errors according to which the number of times each rule of each of the categories of writing faults is not respected is calculated. The errors of writing are the errors of spelling, grammar, conjugation, anglicism, syntax, expression, and use. Thus, this step consists in automatically analyzing the different types of faults appearing in the text. The fourth under-representation represents the stylometry according to which the length of the text, the length of the paragraphs, the length of the sentences, the length of the propositions, the length of the words in characters, the number of each punctuation mark, and finally the parameters of the dialog distribution in the text. The length of the text is calculated from the number of paragraphs, sentences, propositions, words, characters. The length of a paragraph is calculated from the number of sentences, propositions, words, characters. The length of sentences is calculated from the number of propositions, words, characters. The length of the proposals is calculated from the number of words, characters. The calculation of the dialogue distribution parameters in the text is made from criteria chosen from the mean, the variance, the standard deviation, the entropy, the distance between the distributions or a combination of these. Thus, this step identifies the style of the text.

From all the preceding sub-representations, the algorithm generates a fifth under-representation which is a meta-description according to which the vocabulary of the text is analyzed by the different levels of rarity of the words, the lexical fields used, the adapted words to youth, and we calculate aggregations (sums) and ratios (divisions) of the low-level indicators obtained previously.

An example of aggregation calculated from low-level indicators is given below: - NIN = number of verbs in the infinitive - NPR = number of verbs in the present - NFU = number of verbs in the future - NPA = number of verbs in the past. Calculated aggregation yields an intermediate level indicator NV which is the total number of verbs, NV = NIN + NPR + NFU + NPA. An example of a ratio calculated from low-level indicators is given below: - NP = number of sentences - NV = number of verbs.

The calculated ratio gives an intermediate level indicator NM which is the average number of verbs per sentence, ie NM = NV / NP. Thus, this step makes it possible to obtain meta-descriptions such as readability, the extent of the vocabulary or the lexical cohesion.

From the word-bag sub-representation, the algorithm generates a sixth under-representation which represents lexical fields present in the text, by a principal components analysis (PCA). and / or latent semantic analysis (LSA) and / or non-negative Matrix Factorization (NMF). This is therefore a dimensionality reduction step to obtain lexical fields. When one obtains too many lexical fields by these three analyzes, the algorithm generates an additional step of reduction of the dimensionality. This step consists in putting all the lexical fields together and keeping only a small number of them so that those kept are unique and relevant fields. In other words, in case of redundancy in the components of the vector generated according to the preceding step, this step makes it possible to select the non-redundant components of the vector. Once all the vector sub-representations have been generated, the algorithm realizes their concatenation into a final representation of the text.

The algorithm then learns between the components of the vector representation of the learning texts and the vector representations of the groups of readers and the known literary quality of each text of learning, so as to obtain a predictive model of the quality literary according to the groups of readers (step S2).

It is specified that to obtain the known literary quality of each text of learning, we consider the population of reading experts (learning readers). Each reader also gives a mark for each literary text of learning. The notes of a given group obtained by the preceding clustering measure the literary quality of each text for said group and are weighted by being centered and then reduced according to the equation: x '= (x - m) / s where: x is the note given between 1 and 10 by an individual M for a work, m is the average of the notes given by M, s is the standard deviation of the notes given by M, x 'is the new corrected note.

Thus, x 'quantifies the known literary quality of a given group of experts or learning readers. To establish the prediction of the literary quality of a new text the algorithm proceeds as follows. The previously constructed predictive model is applied to the new text and from a given group of readers so as to obtain the literary quality of the new text according to the given group of readers (step S3). The invention just described thus makes it possible to obtain accurately and reliably the literary quality of any text according to a reader profile, the reader profile being determined beforehand by clustering the learning readers.

Many variants and improvements can be envisaged without departing from the scope of the invention.

REFERENCE S CITEES [1]: "DEFT2014, automatic analysis of literary and scientific texts in French", Lecluze et al., 21st Automatic Processing of Natural Languages, Marseille, 2014; [2]: "Fine Semantic Categorization of Expression of Opinion for Consensus Detection", Benamara et al., 21st Automatic Language Processing, Marseille, 2014; [3]: "Revisiting Readability: A Unified Framework for Predicting Text Quality", Pitler et al. In Proceedings of the Conference on Empirical Methods in Natural Language Processing (EMNLP '08). Association for Computational Linguistics, Stroudsburg, PA, USA, 186-195.15

Claims (6)

REVENDICATIONS1. A method of analyzing the literary quality of a text according to the reader profile, implemented by a (micro) computer processor, comprising the steps of: a / receiving a plurality of readers said learning readers; to receive a plurality of texts, called learning texts; b / extract for each learning reader and for the learning texts, its own indicators that are the number of texts read, the ratio between its expression of interest and the number of texts read, and the averages of the notes given to the read texts; c / generate a vector representation of each learning reader from its own indicators; d / subjecting the different vector representations of the learning readers to a data partitioning classifier so as to obtain groups of learning readers; e / generate a vector representation for each group of learning readers by averaging the vector representations of the group's readers; b / extract the numerical indicators of each learning text, c / generate a vector representation of each learning text from its numerical indicators, f / for each group of readers, to learn between the components of the vector representation of the learning texts and vector representations of the groups of readers and the known literary quality of each text of learning, so as to obtain a predictive model of the literary quality according to the groups of readers; g / receive a new text to analyze; h / apply to the new text and from a given group of readers the predictive model built in step d /, so as to obtain the literary quality of the new text according to the given group of readers. 2. Analysis method according to claim 1, comprising, for the vector generation of a learning text according to step c /, the following steps: cl / generating several vector sub-representations of the received text to obtain indicators , so-called low-level indicators, the sub-representations consisting in: - a word bag representation according to which the distributions of each word are analyzed and the distributions of certain unigrams, bi-grams, are analyzed; 3-grams, 4-grams, 5-grams and 6-gram on the scale of the word and characters, - a representation called morphosyntactic structure, according to which one calculates the parameters of the distributions of the grammatical words in the text and one analyzes the distributions of each syntactic function in the text, the paragraphs , sentences and propositions, - a representation of the writing errors according to which one calculates the number of times each rule of each category of writing mistakes is not respected, - a representation of stylometry according to which one computes the length of the text, the length of the paragraphs, the length of the sentences, the length of the propositions, the length of the words in characters, the number of each punctuation mark, and finally the parameters of the distribution of the dialogs in the text; c2 / generate: - a meta-description according to which one analyzes the vocabulary of the text by the different levels of rarity of the words, the lexical fields used, the words adapted to the youth, and one calculates aggregations and ratios of the indicators low-level obtained in 20 cil; a representation of the lexical fields present in the text from the word bag presentation performed in eyelash, by principal component analysis (PCA) and / or a latent semantic analysis (LSA). , acronym for "Latent Semantic Analysis") and / or a non-negative Matrix Factorization (NMF). c3 / concatenation of vector under-representation generated in cl / and c2 /. A computer program comprising program code instructions for executing the steps of the method according to the preceding claim when said program is run on a computer.