Method for retrieving and organizing data originating from patent claims in the field of chemistry or cosmetics

Abstract

A method (100) for processing claims from a patent document, said claims citing chemical entities, said method being implemented by computer and comprising a second step (30-2) including the actions of receiving an annotated set of claims (Ann.Cl) in which a first chemical entity, a second chemical entity, and a verb or verb phrase are identified; detecting the verb or verb phrase; on the basis of the verb or verb phrase, inferring a dependency relationship between the first chemical entity and the second chemical entity; and recording the inferred dependency relationship (DEP) in a dictionary (DIC). Figure to be published with the abstract: Fig. 1

Description

Title of the invention: Method for retrieving and organizing data originating from patent claims in the field of chemistry or cosmetics TECHNICAL FIELD OF THE INVENTION

[0001] The invention lies in the field of Natural Language Processing (NLP), more commonly referred to by the acronym NLP, and more specifically relates to the processing of data consisting of patent claims in the field of chemistry. TECHNOLOGICAL BACKGROUND

[0002] It is common for companies to conduct searches relating to other people's patents or their own patents. For example, companies conduct freedom-to-operate studies to check whether new products they are developing infringe on the patent rights of others. Conversely, the same companies conduct searches to check whether a product that infringes their patents is being marketed by others.

[0003] A patent, a generic term used in this document to refer to a patent application publication or a granted patent, generally comprises three sections: a description, a set of claims, and illustrative figures. The set of claims defines the scope of protection of the patent, and it is therefore this section that defines the subjects relevant for freedom-to-operate studies or for determining whether a product or process falls within the scope of protection of that patent.

[0004] For patent searches, one uses, for example, the patent information platforms of national or regional offices, such as the USPTO (United States Patent and Trademark Office), the EPO (European Patent Office) or the JPO (Japan Patent Office), or paid online search services.

[0005] However, in patent claims, and particularly in patents relating to chemistry or cosmetics, the name of the constituent of a product being searched may not be explicitly stated. Instead, a higher-order encompassing term (e.g., vegetable oil for olive oil) or a term of the same level (e.g., peanut oil for olive oil) serving the same purpose may be found in the claims. Consequently, it is difficult for someone searching for a patent for a product containing a particular constituent to retrieve the relevant patent documents, even if the name of the constituent being searched for is used as a keyword in the search.

[0006] Another source of difficulty lies in the fact that the terms used in the reven The patent document details and the name of the product constituent may be different, and it can be difficult to match a potentially counterfeit product to a patent right.

[0007] US patent application 2022 / 0414160 Al describes the linking of products cited in patent claims and products cited in a product database in order to facilitate the extraction of relevant information, taking into account the different names that may designate the same product.

[0008] However, the methods of extracting information from patent documents still require improvements to enhance the comprehensiveness of searches within such documents and the relevance of the information obtained. Statement of the invention

[0009] An object of the invention is to identify in a set of claims of a patent document chemical entities and chemical properties associated with these chemical entities, and to formalize the links existing between the identified chemical entities, in order to facilitate the analysis of the scope of protection of the patent.

[0010] To achieve this goal, a first aspect of the invention is a method for processing claims from a patent document, said claims citing chemical entities, said method being implemented by computer and comprising a second step including the actions of receiving an annotated set of claims in which a first chemical entity, a second chemical entity, and a verb or verbal phrase are identified; detecting the verb or verbal phrase; on the basis of the verb or verbal phrase, deducing a dependency relationship between the first chemical entity and the second chemical entity; and recording the deduced dependency relationship in a dictionary.

[0011] This process makes it possible to establish a hierarchical relationship between the chemical entities present in the set of claims.

[0012] According to additional, non-limiting features of the first aspect of the invention, considered individually or in any technically feasible combination:

[0013] - a claim number can be recorded in the dictionary which the second chemical entity is cited, in association with this second chemical entity;

[0014] - a grammatical element, a first chemical property and a third entity Chemicals can be identified in the annotated set of claims, the process possibly comprising a third step including actions of detecting a syntactic pattern comprising the grammatical element, the chemical property and the third chemical entity; based on the detected syntactic pattern, a relationship can be deduced of dependence between the third chemical entity and the chemical property; and the deduced dependence relation can be recorded in the dictionary;

[0015] - a claim number can be recorded in the dictionary which chemical property is cited, in association with the third chemical entity;

[0016] - the process may include a fourth step of associating a second chemical property to a fourth chemical entity, including actions of detecting whether a claim in the annotated set of claims has as its chemical entity only the fourth chemical entity and as its chemical property only the second chemical property; and, in response to this detection, a dependency relationship between the fourth chemical entity and the second chemical property can be recorded in the dictionary;

[0017] - the process may include a step of individualizing the claims of the patent document, including a sub-step of identifying dependencies between claims;

[0018] - the process may include a step of identifying chemical entities by com comparison between elements of the claims of the claim set and entries of a computer ontology of the domain of chemistry;

[0019] - the process may include a step of identifying chemical entities and chemical properties by implementing a trained algorithm to detect chemical entities and chemical properties, optimized by machine learning;

[0020] - machine learning can employ a computer ontology of the domain of chemistry;

[0021] - the algorithm may be a so-called Named Entity Recognition algorithm; and

[0022] - the algorithm may employ a so-called Long Short-Term Memory-Conditional model Random Field.

[0023] Additional aspects of the invention relate to:

[0024] - a data processing system comprising means for implementing the steps of the process according to the first aspect of the invention;

[0025] - a computer program comprising instructions which, when the program is executed by a computer, leading the computer to implement the steps of the process according to the first aspect of the invention; and

[0026] - a computer-readable medium comprising instructions which, when they executed by a computer, lead the latter to implement the steps of the process according to the first aspect of the invention. BRIEF DESCRIPTION OF THE FIGURES

[0027] Other features and advantages of the invention will become apparent from the description detailed description of the invention which will follow with reference to the attached figures on which:

[0028] [Fig.1] The [Fig.1] is a diagram representing the whole of a process according to the invention;

[0029] [Fig.2] The [Fig.2] is a diagram representing in detail one of the steps of the process of the [Fig.1];

[0030] [Fig.3] The [Fig.3] represents a diagram of the training of an algorithm implemented in the process of the [Fig.1];

[0031] [Fig.4] The [Fig.4] is a diagram representing in detail one of the steps of the diagram of the [Fig.1];

[0032] [Fig.5] The [Fig.5] illustrates a computer system for implementing the process of the [Fig.1];

[0033] [Fig.6] The [Fig.6] illustrates a set of claims provided to the computer system in step 30 detailed by the [Fig.4];

[0034] [Fig.7] The [Fig.7] illustrates a result at the end of step 30;

[0035] [Fig.8] The [Fig.8] illustrates the application of a first rule of step 30;

[0036] [Fig.9] The [Fig.9] illustrates the application of a second rule from step 30;

[0037] [Fig. 10] Figure 10 illustrates the application of a third rule from step 30; and

[0038] [Fig. 11] The [Fig. 11] illustrates the application of a fourth rule of step 30. DETAILED DESCRIPTION OF THE INVENTION

[0039] The principle of the invention is explained with the aid of figures 1 to 11.

[0040] Figure 1 represents a diagram of the process 100 for processing a PAT patent implemented by a computer system and intended to extract and order the objects contained in its claim set. This process comprises three essential steps.

[0041] A first step 10 consists of identifying the claims of a PAT patent and determining their dependency relationships.

[0042] A second step 20 consists of identifying chemical entities and, where applicable, chemical properties associated with these entities in each of the identified claims. Chemical entities may be identified, for example, by a formula, a common name, or a chemical formula, and may represent simple or complex substances, atoms, molecules, or chemical compounds. A chemical property associated with a chemical entity may be, in particular, a concentration, but more generally any characteristic specifying a chemical property of the chemical entity in question in the claim in which it appears. Examples include viscosity, pH, state (solid, liquid), or form (emulsion, gel) of the chemical entity in question.

[0043] A third step 30 consists of extracting the relationships between each of the entities chemicals identified during step 20 in order to organize the data extracted from the patent into a form easily usable by an end user.

[0044] Step 10 - Individualization of claims

[0045] In step 10, substeps for individualizing claims are applied to the set of claims of a PAT patent received by the system. The claims are identified by applying the rules defined below. Initially, the claims section is identified, for example by locating the section title, "Claim," which must be present in the PAT patent.

[0046] Figure 2 illustrates the sub-steps constituting step 10.

[0047] In a substep 10-1, the claims are separated, for example according to line breaks: in principle, two claims are separated by a line break, which allows them to be automatically distinguished from each other.

[0048] During the processing of a patent application by the offices, it is common for claims to be deleted and indicated as such, for example by the expression "deleted". These are therefore removed from the set of claims for further processing during the second substep 10-2, as being irrelevant to the stated objective, which is to define the subject matter actually contained and defined by the set of claims being analyzed.

[0049] In substep 10-3, the claim numbers are extracted. They can be identified and associated with the corresponding claims by their position at the beginning of the line. A claim without a claim number is associated with a special indication for separate, subsequent processing. Such a situation may arise when the claim set contains an error or when patent manipulations have introduced inaccuracies into the electronic documents actually processed. This can be the case, for example, during character recognition. It can also be the case during automatic translation of a claim set from one language (Chinese, French, German, etc.) to a language to which process 100 has been adapted.

[0050] At a substep 10-4, the title is removed from the set of claims, if applicable.

[0051] In a substep 10-5, a correction is made to the undue separation of a part Regarding the set of claims: it can happen that, due to the identification of a molecule or a value, a single claim has been split into several parts, which are incorrectly considered to be multiple claims. In step 10-5, these parts are detected, for example by the absence of a claim number at the beginning of the line, and merged to recover a single claim instead of a plurality of improperly separated fragments.

[0052] In sub-step 10-6, the claims improperly concatenated, that is to say considered as a single claim when in fact there are a plurality, are separated. This type of situation can occur, for example, when the line break that normally separates two claims was either missing or not detected. This situation can be detected, for example, by a break in the claim numbering sequence.

[0053] In a substep 10-7, potentially missing claim numbers are retrieved. A criterion for determining and correcting such a deficiency may be that the claim numbers must form a continuous sequence of numbers beginning with 1.

[0054] In substep 10-8, the identification of dependencies between claims is carried out. This involves implementing semantic rules on regular expressions to link the claims together. Examples of these regular expressions include the English expressions "preceding daims" or "claim(s) a to b," where a and b represent claim numbers. Thus, claim 10 is associated with claims 1 to 9 if claim 10 contains the expression "preceding daims."

[0055] Substeps 10-1 to 10-8 of step 10 are preferably carried out in that order. Of these, although it is preferable to implement substeps 10-2 and 10-4 to 10-7 in order to improve the performance of the processing, they may be omitted.

[0056] At the end of step 10, the claims were separated and identified in the patent text, and the structure of the set of claims was established, the claims having been individualized and their dependencies on each other having been determined. The claims, which were reworked and annotated, are in Id.Cl form at the end of step 10.

[0057] Step 20 - Detection of chemical entities and their chemical properties

[0058] Step 20 consists of identifying chemical entities and chemical properties in each of the identified claims, preferably on the basis of the Id.Cl claims from step 10. Two alternative approaches are presented in this description.

[0059] A first approach, implemented in step 20A, consists of using a so-called "ontological" database, or computer ontology, defined as a structured corpus of concepts, in this case, data from the field of chemistry. The corpus is modeled in a language that allows a computer to exploit semantic or taxonomic relationships established between these concepts. In the context of the present invention, this involves comparing the claims with existing ontologies in the field of chemistry such as ChEBI (Chemical Entities of Biological Interest), DrugBank, ChEMBL, PubChem, IntEnz (Integrated Relational Enzyme Database), KEGG (Kyoto Encyclopedia of Genes and Genomes), or PDB. (Protein Data Bank).

[0060] According to this first approach, the content of a set of claims and that of one or more selected ontologies are compared. Any element of a claim corresponding to an entry in the ontology or ontologies is considered as a chemical entity and is annotated by means of a tag allowing it to be identified, resulting in a set of claims annotated Ann.ClA.

[0061] A second approach, implemented in step 20B, consists of detecting chemical entities and, preferably, chemical properties based on a so-called "artificial intelligence" technique employing an algorithm optimized by machine learning. For the present invention, this detection falls within the framework of NLP (Natural Language Processing), since it involves extracting information from data presented in textual form: a patent document.

[0062] According to this approach, an algorithmic model is used to process data provided in text form. This data has been defined and then trained using an annotated dataset in order to minimize the average statistical error between an expected response and an actual response during the training phase. The annotated data consists of the same type of data that will be provided to the algorithm during a real implementation phase, but also includes annotations indicating the expected response(s) from the data processing by the algorithm. These annotations allow the calculation of the average statistical error between the expected and actual responses obtained from processing the annotated data. More specifically, this will involve sets of claims comprising chemical entities and chemical properties to be detected, which are already known and annotated as such.The algorithmic model is trained to optimize its ability to detect chemical entities in any set of claims.

[0063] Figure 3 illustrates the 20B-Train training phase of the implemented algorithm. The annotated training data can be Train.Ann.ClA data obtained by successively applying steps 10 and 20A to a corpus of training data Train.Pat. This Train.Ann.ClA data can be directly provided to the algorithm being trained during a step 20B of the algorithm's processing of the provided data, but is preferably further annotated manually during a step 22 interposed between steps 20A and 20B in order to produce the corrected training data Corr.Train.Data with increased relevance compared to the Train.Ann.ClA data. During the training phase, based on the responses returned by the algorithm in a step 20B and by comparison with the annotations provided with the data, a feedback step FB's information system adjusts the algorithm's parameters until a level of reliability deemed sufficient by the user has been reached. It is this algorithm, with its adjusted parameters, that is used in step 20B of process 100 illustrated in [Fig. 1].

[0064] Various algorithmic models can be considered and applied to the problem that the invention aims to solve. However, the inventors determined that a NER (Named Entity Recognition) algorithm employing an LSTM-CRF (Long Short-Term Memory-Conditional Random Field) model for sequence labeling was particularly well-suited to the situation. More specifically, the HunFlair model was used to the inventors' satisfaction. For details on the specifics of HunFlair, see the article by Leon Weber et al. “HunFlair: an easy-to-use tool for state-of-the-art biomedical named entity recognition”, Bioinformatics, Volume 37, Issue 17, September 2021, Pages 2792 to 2794. This model may have been trained ab initio using claim sets annotated to identify the chemical entities and chemical properties sought.It can also be a model already trained to analyze text, for example in English, and then specialized in studying claims including chemical entities and chemical properties, using a process known as fine-tuning. Generally speaking, all methods of model parameterization through "training" will be considered. The claim sets used for training are preferably representative of the claim sets that are intended to be analyzed, in that they include, in particular, chemical entities in the expected forms (chemical formula, non-standard form, etc.) in patents relating to pure chemistry, cosmetics, biology, or related fields.

[0065] Tests on NER algorithms other than HunFlair such as Transformers have shown some efficiency, but less than that of HunFlair for these data.

[0066] Thus, step 20B consists of providing, as input to an algorithm, patents comprising sets of claims in textual form, the claims having been individualized in step 10. The algorithm is preferably a NER algorithm, and among these, preferably an LSTM-CRF algorithm such as the HunFlair algorithm. As output, the algorithm provides a set of claims Ann.ClB annotated with tags designating the identified chemical entities and, preferably, the chemical properties.

[0067] The annotated claim sets Ann.ClA and Ann.ClB are generically indicated as Ann.Cl on the diagram in [Fig.1], one or the other of these two sets being provided for step 30 depending on the approach that will have been used for step 20.

[0068] Step 30 - Extraction of relationships between tagged elements

[0069] Step 30 consists of iteratively populating a DIC dictionary of chemical entities from data extracted from the relationships between the elements—chemical entities and, optionally, chemical properties—identified by tags during step 20, based on the annotated claim set Ann.Cl from step 20. Step 30 can also be designed to increase the overall reliability of the process, taking into account the inherent imperfections in the practical application of textual data processing. As illustrated in [Fig. 4], step 30 comprises six substeps, 30-1 to 30-6, each implementing rules numbered 1 to 6.

[0070] Figure 5 represents a computer system SYS designed to implement process 100 illustrated in Figures 1 to 4. This system comprises a CALC digital computer, an INTF interface for communication between the computer and a user, and a MEM computer memory functionally connected to the computer. The computer system can be distributed, meaning that its elements can be located in multiple places, possibly connected to each other by remote links such as a computer network. When it is explained that a step is implemented, it should be understood that the SYS computer system implements this step using conventional computer techniques, understood by a person skilled in the art.

[0071] Figure 6 illustrates a set of claims such as those provided for processing by step 30. This is a set of claims with annotations indicating chemical entities here by right-angled rectangles and chemical properties by rounded-angle rectangles, the terms or expressions corresponding to these two categories having been identified in step 20.

[0072] In this example, we see that the expression "ph of 8-10" is not indicated, whereas it should be considered a chemical property and indicated as such. This omission is representative of the fact that the processing of data in text form is rarely perfect, and that the determination of a processing algorithm, such as the one defined by step 30, is preferably designed to take imperfections into account, and if possible to correct them.

[0073] Figure 7 illustrates the result that is sought by applying step 30 to the set of claims illustrated in Figure 6: objects, which may be chemical entities or their chemical properties, are identified in the set of claims, associated with the claims in which they are found, and the relationships between these different objects are made formally explicit. This results in a hierarchical organization that can be represented as in (A) of Figure 7, with objects located at nodes N of different levels n, where n = 1, 2, or 3 in the example in Figure 7. Any object at a node of level n+1 specifies an object of level n on which it depends, according to a parent-child dependency relationship. The numbers in parentheses following each element of the representation indicate the claims in which they are found.

[0074] Such a representation can be obtained by assigning to each chemical entity and, optionally, each identified chemical property a "present_in" field comprising the claim(s) in which this chemical entity or chemical property is found, an "info_to_display" field comprising a status of the chemical entity, and a "children" field comprising the chemical entities and chemical properties determined to be dependent on the chemical entity concerned, as illustrated in (B) of [Fig. 7]. [Fig. 7] illustrates in (C) the structure of these fields for the "surfactant" chemical entity of Figures 5 and 6.

[0075] The fields defined in the preceding paragraph also apply to chemical entities dependent on another chemical entity, which leads to a hierarchical structuring of chemical entities on an arbitrary number of node levels.

[0076] It should be noted that the example in Figures 5 and 6 is taken from a set of claims of an actual patent, and reflects the imperfections that can be expected in the processing of textual data, which may be due, for example, to a digitization step including character recognition introducing errors into the text, an application step of artificial intelligence (step 20B for example) also introducing errors, or a step including the use of an incomplete external database (ontology of step 20A for example).

[0077] Step 30 thus implements a series of simple rules detailed below, resulting in the creation of a dictionary listing chemical entities, possible relationships between them, and associating them with chemical properties. Preferably, preparatory steps are carried out prior to step 30.

[0078] A first step, Pre30-l, preparatory to the implementation of step 30, consists of going through the claims one by one and lemmatizing the names of the identified chemical entities. This involves applying a common form to variations (plural, etc.) in the name of the same chemical entity. The following steps are applied to the lemmatized names, or lemmas, obtained, which makes it possible to consider as identical chemical entities represented in the claims of the patent concerned by variations of the same name, variations which may, for example, be due to a plural marker and considered irrelevant for the content analysis of the claims.

[0079] A second step Pre30-2 preparatory to the implementation of step 30 consists of initializing the DIC dictionary by listing in a computer file the chemical entities identified in claim 1 of the claim set during step 20 and associating them in the file at a level called "level 1 node". In addition, each identified chemical entity is associated with claim 1 by indicating the number of that claim in the "present_in" field associated with the entry in the DIC dictionary of that chemical entity.

[0080] A third step, Pre30-3, preparatory to the implementation of step 30, consists of performing morphosyntactic tagging (also called grammatical tagging, or POS-tagging for Part-Of-Speech tagging in English terminology), which involves associating corresponding grammatical information with words in a text using a computer tool. Here, the aim will be to identify nouns, verbs, or coordinating conjunctions and store them in a POS computer file, saved in MEM memory, for later use.

[0081] Step 30 consists of applying a series of rules numbered 1 to 6 during corresponding steps 30-1 to 30-6, preferably executed in that order, to the set of claims annotated Ann.Cl from step 20. The set of claims annotated Ann.Cl will preferably have been submitted to steps Pre30-1 to Pre30-3 before being submitted to step 30.

[0082] A first step 30-1 consists of completing the initialized dictionary by applying rule 1 successively to all claims except claim 1. This rule also has the effect of compensating for possible omissions in the detection of chemical entities during the previous steps.

[0083] Rule 1 consists of adding a chemical entity to the initialized dictionary, associated with a level 1 node, if each of the following three conditions are met: - the chemical entity is present in the text of the first claim; - the chemical entity is not listed in the dictionary; and - the chemical entity is not in the list of ingredients identified in the first claim.

[0084] In addition, each identified chemical entity is associated with the claim in which it appears by indicating the number of that claim in a "present_in" field associated with the entry in the dictionary of that chemical entity.

[0085] Figure 8 illustrates an application of rule 1: when implementing step 30-1, a number of chemical entities (underlined in the text of Figure 6) were identified, but the chemical entity "Chemical filter," although present in the first claim, was only identified in claim 4 and had not yet been added to the dictionary. In response, the entity "Chemical filter" is added to the dictionary and associated with a level 1 node.

[0086] A second step 30-2 consists of using certain verbs or verb phrases present in the claims according to a second rule, so as to define relationships between chemical entities or between a chemical entity and a property chemical relationship. In patents relating to chemical entities, there are verbs often used that convey (or mark) a relationship. This is the case, for example, with the verb "comprising" or the verb phrases "selected from" or "consisting of".

[0087] A list of relationship marker verbs has been prepared in advance and stored in a computer file VER, these verbs being particularly useful for inferring a dependency link between two chemical entities identified in a claim. During step 3-20, the claims are reviewed one by one and, when a verb from the list is detected, for example by detecting a tag placed during the creation of the list of nouns, verbs, or conjunctions in the POS file, the following operations are performed, typically for a textual expression of "IEC being CE2" where IEC and CE2 each represent a chemical entity and "being" is a V verb indicating a DEPI dependency between these two entities, as illustrated by [Fig. 9]:

[0088] - We check for a chemical IEC entity located before the verb V if it is already present in the dictionary;

[0089] - If the IEC chemical entity located before the verb is already present in the dictionary, Only the chemical entity CE2, considered dependent on the chemical entity IEC because it is located after the verb, is added to the dictionary as a chemical entity dependent on the chemical entity IEC. The chemical entity CE2 can, for example, be stored in a field called "children" associated with the chemical entity IEC and comprising all the chemical entities dependent on the chemical entity IEC; and

[0090] - If the IEC chemical entity located before the verb is not already present in the dictionary In addition, a new corresponding entry is created in the dictionary, with the dependent entity CE2 in the "children" field as explained in the preceding paragraph. Furthermore, the entry for entity CE2 in the dictionary includes a "present_in" field in which the claim number mentioning entity CE2 is recorded.

[0091] In a hierarchical representation of the dependency relationships described above, a chemical entity B dependent on a chemical entity A is located at a higher rank node of an increment of 1 with respect to the rank of chemical entity A in a hierarchical representation graph of dependency links.

[0092] Each detected chemical entity is associated with the claim number in which it is detected by recording this number in the "present_in" field of this chemical entity.

[0093] Figure 9 illustrates the application of rule 2, with (A) the chemical entities identified in their respective claims represented by rectangles and verbs or verb phrases identified by dashed ellipses. Figure 8 (B) represents a hierarchical representation of the dependency links between the entities chemical compounds identified based on detected verbs or verb groups.

[0094] In this example, the verbs or verb phrases "being" and "selected from" are detected, allowing the chemical entities located after to be hierarchically linked to the chemical entities located before. Thus, the chemical entity "amphoretic surfactant" is considered dependent on the chemical entity "surfactant" and is therefore classified in the "children" camp of the chemical entity "surfactant," and the claim number in which these two entities are found is associated with them. The chemical entity "surfactant" is located at a level node of n=1, while its dependent chemical entity "amphoretic surfactant" is located at a level node of n+1, i.e., n=2 here. Similarly, it is detected that the four chemical entities "lauryl hydroxy-sultaine," "cocamidopropyl betaine," "coacamidoamphoacetic," and "lauramidopropyl betaine" of claim 5 are dependent on the chemical entity "amphoretic surfactant."These four chemical entities are added to the dictionary in the "children" field of the chemical entity "amphoretic surfactant" and are therefore located at a higher-level node of an increment of 1 at the level of the chemical entity on which they depend, i.e., n=2+l=3 of the hierarchical representation. Claim number 5 is associated with the four chemical entities added to the dictionary as well as with the chemical entity "amphoretic surfactant" on which they depend.

[0095] In the same way as described above, a dependency link is detected by means of the verb "being" between "cl2-c20 fatty acid" on the one hand and "dodecanoic acid", "tetradecanoic acid" and "stearic acid" on the other hand, in claim 3. Also, a dependency link is detected by means of the verbal phrase "selected from" between "inorganic base" on the one hand and "potassium hydroxide" and "sodium hydroxide" on the other hand in claim 5.

[0096] By applying rule 2 successively to all the claims as described above, the dictionary is progressively completed.

[0097] A third step 30-3 consists of applying a third rule to all claims so as to identify syntactic patterns employing grammatical elements that introduce a dependency relationship between chemical entities and chemical properties, and then introducing these chemical properties into the dictionary so as to indicate their respective dependencies on the chemical entities. This step is useful provided that the claims are annotated in such a way that chemical entities as well as chemical properties are marked up. This is, for example, the case when step 20B has been applied to the set of claims.

[0098] The grammatical elements necessary for the implementation of step 30-3 were, for example, identified during the creation of the POS computer file. For example, the preposition "of" in the expression "5-25% by weight of cl2-c20 fatty acids" implies A dependency relationship exists between the chemical property "5-25% by weight" and the chemical entity "cl2-c20 fatty acids". Such a sequence, "chemical property A" + "of" + "chemical entity B", constitutes a syntactic pattern used for dependency detection. A set of such patterns can be prepared in advance and then searched within the set of claims to be analyzed.

[0099] Step 30-3 therefore applies the following operations:

[0100] - Detection of a syntactic pattern and if the pattern is detected; then

[0101] - If the chemical property is not in the dictionary but the entity is in the dictionary, add the chemical property to the dictionary in the "children" field of the chemical entity; and

[0102] - If neither the chemical entity nor the chemical property is in the dictionary, y Add the chemical entity to a level 1 node and add the chemical property to the dictionary in the "children" field of the chemical entity.

[0103] By proceeding in this way, the dictionary is enriched with the chemical properties identified in the claims, these chemical properties being respectively associated with the chemical entities of the dictionary and located at a hierarchical level incremented by 1 with respect to these chemical entities.

[0104] Figure 10 illustrates in (A) chemical entities and chemical properties identified in the claim set already illustrated in [Fig. 9], respectively identified by right-angled rectangles and rounded-angle rectangles. [Fig. 10] illustrates in (B) a hierarchical representation of the dependency links between the chemical entities and the chemical properties identified on the basis of syntactic patterns detected in the claims. In this example, the chemical properties “5-50 wt%”, “5-25% by weight”, and “0.5-9.5 wt%” (PROP1) are respectively associated with the chemical entities “surfactant”, “cl2-c20 fatty acid”, and “inorganic base” (CE3), respectively, by the preposition “of” (PREP). The detected syntactic pattern is PROP1-PREP-CE3 for the expression “0.5-9.5 wt% of an inorganic base”.

[0105] The claim numbers in which these chemical properties are found are also specified in the dictionary. Following this step, chemical properties are considered in the same way as chemical entities, except that a chemical property is necessarily dependent on a chemical entity.

[0106] A fourth step 30-4 consists of applying a fourth rule to all claims except the first. This rule arises from the observation that generally, in a dependent claim, when only one chemical entity and one chemical property are cited, then that chemical property is associated with that chemical entity by adding it to the "children" field of that chemical entity in the dictionary.

[0107] The fourth rule is defined as follows: - If a claim involves only one identified chemical entity; and - If this claim involves only one identified chemical property; then - The identified chemical property is associated with the identified chemical entity.

[0108] Figure 11 illustrates in (A) chemical entities and chemical properties identified in a set of claims, respectively identified by right-angled rectangles and rounded-angle rectangles. Figure 11 illustrates in (B) a hierarchical representation of the dependency links between the chemical entities and the chemical properties identified (i) on the basis of syntactic patterns detected in the claims according to the third rule and (ii) on the basis of the fourth rule defined above.

[0109] In this example, the association of the chemical property "2% in weight" with the chemical entity "film forming agent" was detected using a syntactic pattern rule and added to the dictionary during step 30-3. Similarly, the association of the chemical property "from 0.5 to 7.5% in weight" with the chemical entity "acid phenylbenzimidazolesulfonic" was detected using a syntactic pattern rule and added to the dictionary. In contrast, the associations between the chemical properties "going from 1 to 5% in weight" (PROP2) and "in a quantity of 3% in weight" and the chemical entity "acid phenylbenzimidazolesulfonic" (CE4) were detected by implementing the fourth rule during step 30-4.

[0110] As with the application of the third rule, in addition to the dependence relationship with respect to the chemical entity it characterizes, the claim number containing this dependence relationship is associated in the dictionary with the identified chemical property. In the example of [Fig. 1 1], we see in (B) that claims 1, 2 and 3 respectively contain the three chemical properties "from 0.5 to 7.5% in weight", "going from 1 to 5% in weight" and "in a quantity of 3% in weight" associated with the chemical entity "acid phenylbenzimidazolesulfonic".

[0111] A fifth step 30-5 aims to integrate into the dictionary chemical entities that may have been omitted during the previous steps.

[0112] During step 30-5, the following operations are carried out:

[0113] - Claim by claim, it is verified whether the chemical entities identified in The set of claims are listed in the dictionary;

[0114] - If an identified chemical entity is not present in the dictionary, then it is added there adds in such a way as to form a level 1 node; and

[0115] - If a chemical entity is present in the dictionary but the number of a If the claim in which this chemical entity is present is not specified in the "present_in" field, then this claim number is added in this field.

[0116] A sixth step 30-6 consists of applying a sixth rule to the set of re claims, allowing for the inclusion of coordinating conjunctions such as "and" and "or," and exclusions such as "not" and "free of." This step adds relevant information to the interpretation of the claims in the DIC dictionary and displays it, where appropriate, in a hierarchical representation of dependency links between graphs.

[0117] In practice, following the same principle as in Rule 2, syntactic patterns are identified and used to enrich the dictionary. In this case, these syntactic patterns include coordinating conjunctions listed in the POS file, and their detection allows for a more precise definition of the link between a chemical entity and the objects defined by the claim set. More specifically, each coordinating conjunction is assigned a status for the associated chemical entity. This status will be integrated into the dictionary in association with that chemical entity.

[0118] During step 30-6, the following operations are carried out:

[0119] - A syntactic pattern comprising a is sought in the claims coordinating conjunction listed in the POS file and a chemical entity; and

[0120] - if such a syntactic pattern is identified, then the status assigned to the coordinating conjunction The ordination of the identified syntactic pattern is associated with the chemical entity in the dictionary, for example by recording in the "info_to_display" field.

[0121] The content of the "info_to_display" field can, where appropriate, be displayed on the hierarchical representation deduced from the dictionary.

[0122] The example in Figures 5 and 6 illustrates the addition of the "freeof" status to the detected chemical entity "thickener," indicating that the subject matter of the claim must not contain "thickener." Similarly, the "and, or" status is added to the chemical entities "moisurizing agent," "preservative," and "emollient" in claim 7, indicating that the subject matter of the claim may include one or more of these chemical entities.

[0123] Of course the invention is not limited to the embodiments described above and alternative embodiments can be made without departing from the scope of the invention as defined by the claims.

Claims

Claims

1. A method (100) for processing claims of a patent document, said claims citing chemical entities, said method being computer-implemented and comprising a second step (30-2) comprising the actions of: - receiving an annotated set of claims (Ann.Cl) in which a first chemical entity (CEI), a second chemical entity (CE2), and a verb or verbal phrase (V) are identified; - detecting the verb or verbal phrase (V); - based on the verb or verbal phrase (V), deducing a dependency relationship (DEP) between the first chemical entity (CEI) and the second chemical entity (CE2); and - recording in a dictionary (DIC) the deduced dependency relationship (DEP).

2. The method (100) of claim 1, wherein a number of a claim in which the second chemical entity (CE2) is cited is recorded in the dictionary, in association with this second chemical entity.

3. Method (100) according to claim 1 or 2, wherein a grammatical element (PREP), a first chemical property (PROP1) and a third chemical entity (CE3) are identified in the annotated set of claims (ANN.C1), the method further comprising a third step (30-3) comprising the actions of: - detecting a syntactic pattern (PROP1-PREP-CE3) comprising the grammatical element (PREP), the chemical property (PROP1) and the third chemical entity (CE3); - based on the detected syntactic pattern, deducing a dependency relationship (DEP2) between the third chemical entity (CE3) and the chemical property (PROP1); and - recording in the dictionary (DIC) the deduced dependency relationship (DEP2).

4. A method (100) according to any one of claims 1 to 3, wherein a number of a claim in which the chemical property (PROP1) is cited is recorded in the dictionary, in association with the third chemical entity (CE3).

5. Method (100) according to any one of claims 1 to 4, further comprising a fourth step (30-4) of associating a second chemical property (PROP2) to a fourth chemical entity (CE4), comprising the actions of: - detecting whether a claim in the annotated set of claims (Ann.Cl) only includes as a chemical entity the fourth chemical entity (CE4) and only the second chemical property (PROP2) as a chemical property; and - in response to this detection, recording in the dictionary (DIC) a dependency relationship between the fourth chemical entity (CE4) and the second chemical property (PROP2).

6. Method according to any one of claims 1 to 5, further comprising a step (10) of individualizing the claims of the patent document, comprising a sub-step (10-8) of identifying dependencies between claims.

7. A method according to any one of claims 1 to 6, further comprising a step (20A) of identifying chemical entities by comparison between elements of the claims of the claim set and entries of a computer ontology in the field of chemistry.

8. A method according to any one of claims 1 to 7, further comprising a step (20B) of identifying chemical entities and chemical properties by implementing a trained algorithm for detecting chemical entities and chemical properties, optimized by machine learning (20B-Train).

9. The method of claim 8, wherein the machine learning employs a computer ontology from the field of chemistry.

10. The method according to claim 8 or 9, the algorithm being a so-called Named Entity Recognition algorithm.

11. The method of claim 10, the algorithm employing a so-called Long Short-Term Memory-Conditional Random Field model.

12. Data processing system comprising means for implementing the steps of the method according to any one of claims 1 to 11.

13. A computer program comprising instructions which, when the program is executed by a computer, cause the computer to implement the steps of the method according to any one of claims 1 to 11.

14. A computer-readable medium comprising instructions which, when executed by a computer, cause the computer to carry out the steps of the method according to any one of the re- claims 1 to 11.

Images