Method and system for multiparametric processing of information from multidimensional and / or multitemporal datasets

Abstract

The present invention relates to a method and a system for the multiparametric processing of information from multidimensional and / or multitemporal datasets. According to the invention, such a method comprises the following four macrostages: a) macrostage (A) of input and pre-processing of data relating to an entity; b) macrostage (B) of processing said data by means of a neural network-based artificial intelligence algorithm; c) macrostage (C) of training said neural network; d) macrostage (D) of indexing and querying said data processed by said macrostage (B) of processing.

Description

[0001] “METHOD AND SYSTEM FOR MULTIPARAMETRIC PROCESSING OF INFORMATION FROM MULTIDIMENSIONAL AND / OR MULTITEMPORAL DATASETS”

[0002] * * *

[0003] FIELD OF THE INVENTION

[0004] The present invention relates to a method and a system for multiparametric processing of information from multidimensional and / or multitemporal datasets relating to a same entity under observation, by means of data processing, synthesis, and classification techniques that employ cognitive artificial intelligence algorithms.

[0005] In particular, the present invention finds specific application in the field of human resources, where the entity under observation may be an individual, such as a candidate for a job offer, or a plurality of individuals, such as a group of employees to be arranged into an optimal work team configuration. The invention is also applicable in the field of medical record analysis, where the entity under observation may be a single patient, or a plurality of patients sharing, for example, the same pathology, and in general in all fields in which current and historical data relating to a certain entity (or subject, or object) must be analyzed in order to understand its classification (correspondence) and / or infer its future significance (probability).

[0006] BACKGROUND OF THE INVENTION

[0007] As is known, in the general field of artificial intelligence, systems must be capable of acquiring information from the surrounding environment, learning its characteristics through the processing of information from multiple sources and datasets, and using this information to highlight elements, structures, or events that are not immediately visible, in relation to a defined time interval, for a past purpose (historical data analysis), a present purpose (data analysis regardless of the time interval), or a future purpose (predictive data analysis).

[0008] Currently, however, there are several areas in which such artificial intelligence systems are not yet adequately applied.

[0009] An example is described in patent EP2503477B1, which discloses a method and system for the contextual search and retrieval of curricula vitae based on attributes derived from curricula stored in a repository. This system includes various steps, such as preprocessing of the curricula, indexing of parameters and derived attributes, formulation of search queries, searching within the index, assigning scores to the identified parameters, converting the initial scores into overall scores, and displaying the relevant results.

[0010] However, systems of the type described in the aforementioned patent suffer from several significant drawbacks. One of the main problems concerns the relevance of the results: keyword- based searches often return irrelevant results due to the lack of adequate context. Furthermore, data non-uniformity also leads to considerable issues. In fact, data submitted by users are often heterogeneous and originate from different sources, requiring extensive preprocessing — generally performed manually — in order to be used effectively. In addition, traditional systems are often neither scalable, thus encountering difficulties in handling increasing data volumes, nor easily adaptable to new application domains, thereby limiting their effectiveness in dynamic and evolving contexts.

[0011] SUMMARY OF THE INVENTION

[0012] In light of the foregoing, the object of the present invention is to provide a method and a system for processing information from multidimensional, multiparametric and / or multitemporal datasets (e.g., textual documents, video, images, audio recordings) that overcomes the limitations of the prior art by enabling fast operation and by reliably and securely highlighting, through the execution of computational operations, information that is processed, synthesized and / or newly generated and classified according to parameters of the surrounding environment.

[0013] Within this context, an aim of the present invention is to provide a method and a system capable of effectively operating starting from a plurality of sources and / or a plurality of formats from which to process data related to different parameters, also acquired at different time instances, thus enabling the generation of multiple classifications of the processed data.

[0014] A further object of the invention is to provide a method and a system capable of ensuring the highest reliability and security in use.

[0015] Another object of the invention is to provide a method and a system that are easy to implement and economically competitive when compared to the prior art.

[0016] The above-stated aim, as well as the mentioned and other objects that will become more apparent hereinafter, are achieved by a method for multiparametric processing of information from multidimensional and / or multitemporal datasets comprising the following four macrostages: a) macrostage (A) of input and pre-processing of data relating to an entity; b) macrostage (B) of processing said data by means of a neural network-based artificial intelligence algorithm; c) macrostage (C) of training said neural network; d) macrostage (D) of indexing and querying said data processed by said macrostage (B) of processing; in which said macrostage (A) of input and pre-processing of data comprises at least the following phases: phase (la) of collecting data relating to an entity, in which a plurality of data from one or more data sources is collected, said data comprising at least text documents, images, videos and audio recordings; a phase (2a) of pre-processing and converting said data into text, in which said text documents, images, videos and audio recordings are converted into textual formats; phase (4a) of transforming said data in which said data are extracted and transformed in a format adapted to be processed by said neural network-based artificial intelligence algorithm; phase (5a) of storing said data, in which said data are stored in a repository; said macrostage (C) of training said neural network comprises at least the following phases: phase (1c) of collecting training data for training said neural network, in which training data are collected from a plurality of sources; phase (2c) of training said neural network, in which supervised and / or unsupervised learning is performed to train said neural network; phase of generating said neural network comprising a phase (4c) of generating a customized neural network, in which a customized neural network is implemented for a specific user and / or for a specific class of users; said macrostage (B) of processing comprises at least the following phases: phase (lb) of processing said pre-processed data in said macrostage (A) of input and pre-processing of data, in which said pre-processed data are processed by means of a neural network-based artificial intelligence algorithm based on said neural network generated in said phase (4c) of generating said neural network and in which new data associated with said pre-processed data are generated; phase (14b) of merging said data, in which corresponding data from different sources are merged; said macrostage (D) of indexing and querying comprises the following phases: phase (Id) of merging said data relating to said entity and said new data, in which input data relating to said entity are merged with said new data generated in said macrostage (B) of processing and in which merged data are generated and stored in said repository; phase (2d) of indexing the merged data, in which said merged data are indexed according to one or more indexes and indexed data are generated; phase (4d) of collecting queries from a user, in which the insertion of one or more search queries by said user is allowed; phase (5d) of processing said one or more queries, in which said one or more search queries are processed; phase (6d) of outputting said one or more queries, in which the results of said one or more search queries are filtered and visualized.

[0017] The above-stated aim and objects are also achieved by a system for multiparametric processing of information from multidimensional and / or multitemporal datasets comprising means for executing the method above, said means comprising the following four macrostages: a) a macrostage (A) of input and pre-processing of data related to an entity; b) a macrostage (B) of processing said data through an artificial intelligence algorithm based on a neural network; c) a macrostage (C) of training said neural network; d) a macrostage (D) of indexing and querying said data processed by said macrostage (B) of processing; in which said macrostage (A) of input and pre-processing of data comprises at least the following modules: module (la) of collecting data related to an entity, configured to collect a plurality of data from one or more data sources, said data comprising at least textual documents, images, videos and audio recordings; module (2a) of pre-processing and converting said data into text, in which said textual documents, images, videos and audio recordings are converted into textual formats; module (4a) of transforming said data in which said data are extracted and transformed in a format adapted to be processed by said neural network-based artificial intelligence algorithm; module (5a) of storing said data, configured to store said data in a repository; said macrostage (C) of training said neural network comprises at least the following modules: module (1c) of collecting training data for training said neural network, configured to collect training data from a plurality of sources; module (2c) of training said neural network, configured to perform supervised and / or unsupervised learning to train said neural network; module of generating said neural network comprising a module (4c) of generating a personalized neural network, configured to implement a personalized neural network for a specific user and / or for a class of specific users; said macrostage (B) of processing comprises at least the following modules: module (lb) of processing said pre-processed data from said macrostage (A) of input and pre-processing of data, configured to process said pre-processed data through an artificial intelligence algorithm based on said neural network generated in said module (4c) of generating said neural network and to generate new data associated with said pre-processed data; module (14b) of merging said data, configured to merge corresponding data from different sources; said macrostage (D) of indexing and querying comprises the following modules: module (Id) of merging said data related to said entity and said new data, configured to merge the input data related to said entity with said new data generated by said macrostage (B) of processing and to generate merged data, stored in said repository; module (2d) of indexing the merged data, configured to index said merged data according to one or more indices and to generate indexed data; module (4d) of collecting queries from a user, configured to allow the insertion of one or more search queries by said user; module (5d) of processing said one or more queries, configured to process said one or more search queries; module (6d) of outputting said one or more queries, configured to filter and display the results of said one or more search queries.

[0018] Additional features are provided in the dependent claims.

[0019] LIST OF THE FIGURES

[0020] Further features and advantages will become more apparent from the following description, provided by way of example and without limitation, of a preferred embodiment of the present invention, illustrated with the aid of the flow diagram shown in Figure 1.

[0021] DETAILED DESCRIPTION OF THE INVENTION

[0022] With particular reference to the flow diagram of Figure 1, the method for multiparametric processing of information from multidimensional and / or multitemporal datasets, implemented on a computer, comprises the following four macrostages: a) macrostage (A) of input and pre-processing of data relating to an entity; b) macrostage (B) of processing said data by means of a neural network-based artificial intelligence algorithm; c) macrostage (C) of training said neural network; d) macrostage (D) of indexing and querying said data processed by said macrostage (B) of processing.

[0023] The term entity refers to the object, the subject, or a plurality thereof, to which the data processed by the system are associated and which depends on the type of application for which the system is implemented.

[0024] For example, in the case of application in the field of human resources, the entity may be a worker, or a group of workers, whereas in the case of application in the healthcare field, for medical record analysis, the entity may be a patient, or a group of patients, for instance sharing the same pathology.

[0025] According to the invention, each of the four macrostages (A-D) comprises a plurality of phases as described below, with reference to the flow diagram of Figure 1.

[0026] Macrostage (A) of input and pre-processing of data comprises at least the following phases: phase (la) of collecting data relating to an entity, in which a plurality of data from one or more data sources is collected; phase (5a) of storing said data, in which said data are stored in a repository, namely a digital archive, preferably centralized, in which the data are stored and managed in a structured manner.

[0027] In macrostage (C), the neural network (or a plurality of coordinated neural networks) to be used in macrostage (B) for data processing is configured.

[0028] Macrostage (C) of training said neural network comprises at least the following phases:

[0029] - phase (1c) of collecting training data for training said neural network, in which training data are collected from a plurality of sources;

[0030] - phase (2c) of training said neural network, in which supervised and / or unsupervised learning is performed to train said neural network;

[0031] - phase (3c, 4c) of generating said neural network, comprising at least one of: a phase (3c) of generating a basic neural network, in which a basic neural network is implemented for a generic user, and a phase (4c) of generating a customized neural network, in which a customized neural network is implemented for a specific user and / or for a specific class of users.

[0032] Macrostage (B) of processing comprises at least the following phases:

[0033] - phase (lb) of processing said pre-processed data from said macrostage (A) of input and pre-processing of data, in which said pre-processed data are processed by means of a neural network-based artificial intelligence algorithm generated in said phase (3c, 4c) of generating said neural network, and in which new data associated with said pre- processed data are generated;

[0034] - phase (14b) of merging said data, in which corresponding data from different sources are merged.

[0035] Finally, macrostage (D) of indexing and querying comprises the following phases:

[0036] - phase (Id) of merging said data relating to said entity and said new data, in which the input data relating to said entity are merged with said new data generated in said macrostage (B) of processing, and in which merged data are generated and stored in the aforementioned repository;

[0037] - phase (2d) of indexing the merged data, in which said merged data are indexed according to one or more indices and indexed data are generated;

[0038] - phase (4d) of collecting queries from a user, in which the insertion of one or more search queries by said user is allowed;

[0039] - phase (5d) of processing said one or more queries, in which said one or more search queries are processed;

[0040] - phase (6d) of outputting said one or more queries, in which the results of said one or more search queries are filtered and displayed.

[0041] Preferably, macrostage (A) of input and pre-processing of data comprises one or more of the following phases:

[0042] - a phase (2a) of pre-processing and converting said data into text, in which said data are converted into textual formats;

[0043] - a phase (3a) of analyzing said data when collected in the form of a textual document, in which a document layout analysis (DLA) is applied to identify and separate different sections present in said document;

[0044] - a phase (4a) of transforming said data, in which text is extracted from videos and in which said data are translated into a selected language;

[0045] - a phase (6a) of applying templates for parsing analysis, in which one or more templates are applied to said data to facilitate the parsing of said data, by classifying and prioritizing identified sections in said data, thereby obtaining pre-processed data.

[0046] Preferably, phase (2a) occurs before phase (3a). Preferably, phase (2a), phase (3a), and phase (4a) are carried out between phase (la) and phase (5a) of macrostage (A).

[0047] Preferably, macrostage (B) of processing comprises one or more of the following phases:

[0048] - a phase (2b) of parallel control, in which an analysis is performed using third-party artificial intelligence algorithms and in which additional information from said third- party artificial intelligence algorithms is incorporated into said data;

[0049] - a phase (3b) of controlled integration of said data, in which additional information from controlled data sources, such as manual sources, is incorporated into said data;

[0050] - a phase (4b) of automated importation of said additional information, in which said additional information is automatically imported into said data through an API (Application Programming Interface);

[0051] - a phase (5b) of manual compilation of forms, in which manual insertion of information for the purpose of integrating said data is allowed;

[0052] - a phase (6b) of classifying said new data, in which new classifications of said new data are generated by classifying them at multiple levels;

[0053] - a phase (7b) of reclassifying said new data, in which said new data are reclassified so as to associate them with said entity using unique identifiers of said entity;

[0054] - a phase (8b) of generating summaries and properties, in which summaries of said new data are generated and in which descriptive elements of the properties of said entity are generated;

[0055] - a phase (9b) of extracting keywords, in which keywords and / or groups of keywords are extracted and classified from said new data;

[0056] - a phase (10b) of generating new information, in which new information is generated based on said new data;

[0057] - a phase (11b) of generating a proximity percentage, in which the proximity percentage is calculated between said new data and the information present in said phase (3c, 4c) of generating said neural network and / or in said phase (8c) of collecting sources for the preparation of said knowledge semantic graphs of said phase (7c) of preparing knowledge semantic graphs and / or in said phase (9c) of enriching said neural network;

[0058] - a phase (12b) of visualizing said knowledge semantic graphs, in which graphical and textual outputs of said knowledge semantic graphs applied to said new data are generated;

[0059] - a phase (13b) of identifying implications and correlations, in which implications and correlations between said new data are identified.

[0060] Preferably, phase (2b), phase (3b), phase (4b), phase (5b), phase (6b), phase (7b), phase (8b), phase (9b), phase (10b), phase (11b), phase (12b), and phase (13b) are carried out between phase (lb) and phase (14b) of macrostage (B).

[0061] Preferably, phase (2b) and phase (3b) may be carried out in parallel with each other and / or in parallel with phase (lb). Preferably, phase (6b), phase (7b), phase (8b), phase (9b), and phase (10b) may be carried out in parallel with one another, and / or all in parallel with each other.

[0062] Preferably, macrostage (C) of training said neural network comprises one or more of the following phases:

[0063] - a phase (5c) of preparing basic knowledge models, in which basic knowledge models are prepared, usable as additional sources of training data for the generation of said neural network in said phase (3c) of generating a basic neural network and / or in said phase (4c) of generating a personalized neural network;

[0064] - a phase (6c) of preparing basic dictionaries, in which dictionaries are prepared, usable as additional sources of training data for the generation of said neural network in said phase (3c) of generating a basic neural network and / or in said phase (4c) of generating a personalized neural network;

[0065] - a phase (7c) of preparing knowledge semantic graphs, in which semantic graphs are prepared, usable as additional sources of training data for the generation of said neural network in said phase (3c) of generating a basic neural network and / or in said phase (4c) of generating a personalized neural network;

[0066] - a phase (8c) of collecting sources for the preparation of said knowledge semantic graphs of said phase (7c) of preparing knowledge semantic graphs, in which implications and correlations between the elements of said knowledge semantic graphs are generated;

[0067] - a phase (9c) of enriching said neural network, in which said neural network generated in said phase (3c, 4c) of generating said neural network is enriched with information generated at least in said phase (lb) of processing said pre-processed data and in one or more of: said phase (5c) of preparing basic knowledge models, said phase (6c) of preparing basic dictionaries, and said phase (7c) of preparing knowledge semantic graphs.

[0068] Preferably, phase (5c), phase (6c), and phase (7c) are carried out after phase (3c, 4c) of generating a neural network in macrostage (C). Preferably, phase (8c) precedes phase (7c).

[0069] Preferably, in said phase (9c) of enriching said neural network, said neural network generated in said phase (3c, 4c) of generating said neural network is enriched with information originating not only from said phase (lb) of processing said pre-processed data but also from said phase (2b) of parallel control and / or from said phase (3b) of controlled integration of said data.

[0070] Preferably, macrostage (D) of indexing and querying comprises one or more of the following phases: - a phase (3d) of ranking said indexed data, in which a ranking for each of said one or more indices is assigned to said indexed data;

[0071] - a phase (7d) of manual manipulation of said new data generated in said macrostage (B) of processing, in which said user is allowed to enrich, correct, and modify said new data generated in said macrostage (B) of processing;

[0072] - a phase (8d) of reusing said manipulated new data in said phase (7d) of manual manipulation, in which said manipulated new data is used as an additional source of training data for generating said neural network in said phase (3c) of generating a basic neural network and / or in said phase (4c) of generating a personalized neural network, and in which further training cycles of said neural network are initiated based on said phase (2c) of training said neural network.

[0073] Advantageously, as also illustrated in the flow diagram of Figure 1, macrostage (B) and macrostage (C) may be executed at least in part concurrently and in parallel. In fact, some modules of macromodule (B) interact with and exchange data and information with certain modules of macromodule (C), and vice versa.

[0074] The present invention also relates to a computer program comprising instructions which, when the program is executed by a computer, cause said computer to carry out the above-described method. Said computer program is preferably loaded onto said repository.

[0075] The present invention also relates to a system for multiparametric processing of information from multidimensional and / or multitemporal datasets, comprising means, such as for example a computer, for executing the above-described method.

[0076] In particular, the system for multiparametric processing of information from multidimensional and / or multitemporal datasets comprises the following four macromodules: a) macromodule (A) of input and pre-processing of data relating to an entity; b) macromodule (B) of processing said data by means of a neural network-based artificial intelligence algorithm; c) macromodule (C) of training said neural network; d) macromodule (D) of indexing and querying said data processed by said macromodule (B) of processing.

[0077] According to the invention, each of the four macromodules (A-D) comprises a plurality of modules, as described below with reference to the flow diagram of Figure 1.

[0078] Each macromodule (A-D) of the system corresponds to a respective macrostage (A-D) of the method, and each module of the system is preferably configured to execute a corresponding phase of the method according to the invention. Macromodule (A) of input and pre-processing of data comprises at least the following modules: module (la) of collecting data relating to an entity, configured to collect a plurality of data from one or more data source; module (5a) of storing said data, configured to store said data in a repository.

[0079] Macromodule (C) of training said neural network comprises the following modules: module (1c) of collecting training data for training said neural network, configured to collect training data from a plurality of sources; module (2c) of training said neural network, configured to perform supervised and / or unsupervised learning to train said neural network; module (3c, 4c) of generating said neural network, comprising at least one of a module (3c) of generating a basic neural network, configured to implement a basic neural network for a generic user, and a module (4c) of generating a personalized neural network, configured to implement a personalized neural network for a specific user and / or for a specific class of users.

[0080] Macromodule (B) of processing comprises at least the following modules: module (lb) of processing said pre-processed data from said macromodule (A) of input and pre-processing of data, configured to process said pre-processed data by means of an artificial intelligence algorithm based on the neural network generated by said module (3c, 4c) of generating said neural network, and to generate new data associated with said pre-processed datai; module (14b) of merging data, configured to merge corresponding data from different sources.

[0081] Finally, macromodule (D) of indexing and querying comprises the following modules: module (Id) of merging said data relating to said entity and said new data, configured to merge the input data relating to said entity with said new data generated by said macromodule (B) of processing, and to generate merged data: module (2d) of indexing the merged data, configured to index said merged data according to one or more indices and to generate indexed data; module (4d) of collecting queries from a user, configured to allow the insertion of one or more search queries by said user; module (5d) of processing said one or more queries, configured to process said one or more search queries; module (6d) of outputting said one or more queries, configured to filter and display the results of said one or more search queries.

[0082] Macromodule (A) further comprises a module (2a) of pre-processing and converting data into text, configured to convert said data into textual formats.

[0083] Preferably, macromodule (A) further comprises a module (3a) of analyzing data when collected in the form of a textual document, configured to apply document layout analysis (DLA) to identify and separate different sections present in said document, such as, for example, work and education experiences.

[0084] Macromodule (A) further comprises a module (4a) of transforming data, preferably configured to extract text from videos and to translate the data into a selected language.

[0085] Preferably, macromodule (A) comprises a module (6a) of applying templates for parsing analysis, configured to apply one or more templates to said data to facilitate parsing of said data, by classifying and prioritizing identified sections in said data, thereby obtaining pre-processed data.

[0086] Macromodule (A) may also comprise a combination of two or more of said modules (2a, 3 a, 4a, and 6 a).

[0087] Preferably, macromodule (B) further comprises a module (2b) of parallel control, configured to integrate the analysis with third-party artificial intelligence algorithms and to incorporate into said data additional information from said third-party artificial intelligence algorithms.

[0088] Preferably, macromodule (B) further comprises a module (3b) of controlled integration of said data, configured to incorporate into said data additional information from controlled data sources.

[0089] Preferably, macromodule (B) further comprises a module (4b) of automated importation of additional information, configured to automatically import said additional information into the data through an application programming interface (API).

[0090] Preferably, macromodule (B) further comprises a module (5b) of manual compilation of forms, configured to allow the manual insertion of information for the purpose of integrating the data.

[0091] Preferably, macromodule (B) further comprises a module (6b) of classification of said new data, configured to generate new classifications of said new data by classifying them at multiple levels. Preferably, macromodule (B) comprises a module (7b) of reclassification of said new data, configured to reclassify said new data so as to associate them with said entity using unique identifiers of said entity.

[0092] Preferably, macromodule (B) comprises a module (8b) of generating summaries and properties, configured to generate summaries of said new data and to generate descriptive elements of the properties of said entity.

[0093] Preferably, macromodule (B) comprises a module (9b) of keyword extraction, configured to extract and classify from said new data keywords and / or groups of keywords.

[0094] Preferably, macromodule (B) comprises a module (10b) of generating new information, configured to generate new information based on said new data.

[0095] Macromodule (B) may also comprise a combination of two or more of said modules (4b, 5b, 7b, 8b, 9b, 10b).

[0096] Preferably, macromodule (B) further comprises a module (11b) of generating a proximity percentage, configured to calculate the proximity percentage between said new data and the information contained in said module (3c, 4c) of generating said neural network and / or in said module (8c) of collecting sources for the preparation of said knowledge semantic graphs of said module (7c) of preparing knowledge semantic graphs and / or in said module (9c) of enriching said neural network.

[0097] Preferably, macromodule (B) further comprises a module (12b) of visualizing said knowledge semantic graphs, configured to generate graphical and textual outputs of said knowledge semantic graphs applied to said new data.

[0098] Preferably, macromodule (B) further comprises a module (13b) of identifying implications and correlations, configured to identify implications and correlations between said new data.

[0099] Preferibilmente, il macromodulo (C) comprende ulteriormente un modulo (5c) di predisposizione di modelli di conoscenza di base, configurate per predisporre modelli di conoscenza di base utilizzabili come ulteriori fonti di dati di training per la generazione di detta rete neurale in detto modulo (3c) di generazione di una rete neurale di base e / o in detto modulo (4c) di generazione di una rete neurale personalizzata;

[0100] Preferibilmente, il macromodulo (C) comprende un modulo (6c) di predisposizione di dizionari di base, configurate per predisporre dizionari utilizzabili come ulteriori fonti di dati di training per la generazione di detta rete neurale in detto modulo (3c) di generazione di una rete neurale di base e / o in detto modulo (4c) di generazione di una rete neurale personalizzata.

[0101] Preferably, macromodule (C) further comprises a module (5c) of preparing basic knowledge models, configured to prepare basic knowledge models usable as additional sources of training data for generating said neural network in said module (3c) of generating a basic neural network and / or in said module (4c) of generating a personalized neural network.

[0102] Preferably, macromodule (C) comprises a module (8c) of collecting sources for the preparation of said knowledge semantic graphs of said module (7c) of preparing knowledge semantic graphs, configured to generate implications and correlations between the elements of said knowledge semantic graphs.

[0103] Preferably, macromodule (C) comprises a module (9c) of enriching said neural network, configured to enrich said neural network generated by said module (3c, 4c) of generating said neural network with information originating at least from said module (lb) of processing said pre- processed data and from one or more of: said module (5c) of preparing basic knowledge models, said module (6c) of preparing basic dictionaries, and said module (7c) of preparing knowledge semantic graphs.

[0104] Preferably, module (9c) is configured to enrich the neural network generated by module (3c, 4c) of generating said neural network with information originating not only from module (lb), but also from module (2b) and / or module (3b).

[0105] Preferably, macromodule (D) further comprises a module (3d) of ranking said indexed data, configured to assign a ranking for each of said one or more indices to said indexed data.

[0106] Preferably, macromodule (D) comprises a module (7d) of manual manipulation of said new data generated by said macromodule (B) of processing, configured to allow said user to enrich, correct, and modify said new data generated by said macromodule (B) of processing.

[0107] Preferably, macromodule (D) comprises a module (8d) of reusing said manipulated new data in said module (7d) of manual manipulation, configured to use said manipulated new data as an additional source of training data for generating said neural network in said module (3c) of generating a basic neural network and / or in said module (4c) of generating a personalized neural network, and to initiate further training cycles of said neural network through said module (2c) of training said neural network.

[0108] Preferably, module (la) collects the plurality of data from one or more sources, such as curricula vitae, Linkedln profdes, personnel records in the company’s human resources management system, and video interviews in the case of application in the employment sector, or sources such as medical records, wearable systems (e.g., watches, sensorized rings, sensorized bands), medical repositories, as well as medical / scientific portals and / or journals (e.g., PubMed). In particular, the plurality of data from each source is processed by microservices that extract and forward such data to the subsequent modules. Preferably, the templates applied to the data in module (6a) may include templates specifically designed ad hoc or templates imported through external libraries.

[0109] The operations performed by the system for multiparametric processing of information from multidimensional and / or multitemporal datasets, according to the method described herein, are explained below with reference to the flow diagram of Figure 1.

[0110] In particular, with regard to macromodule A, module la for collecting data relating to an entity contains the starting sources for the cognitive artificial intelligence process. These sources may be considered individually or as a summation.

[0111] Each source serves as the input for a specific microservice, which operates ad hoc to extract the information and forward it to the analysis process.

[0112] The microservices may include:

[0113] - Curriculum (or medical record) analysis: the file is uploaded by the microservice or sent via the API, either individually or as part of a bulk upload involving multiple files. The analysis process is managed through queues and priority levels;

[0114] - Analysis of professional social profiles (or recordings from wearable systems and / or medical repositories): starting from the link to the social profile — or from a social profile link extracted from the corresponding curriculum — the microservice performs scraping of public information from the social network, particularly text scraping and profile download, using functionalities made available by the social platform itself, and forwards the textual information to the artificial intelligence process;

[0115] - Analysis of the personnel file (or medic al / scientific journals and portals): in cases where a personnel file is available, for example in an enterprise Human Management System software, including personal data and additional information such as the projects the employee has worked on, the microservice accesses such file;

[0116] - Possibility to analyze other sources using the same procedure, such as video interviews (or audio recordings of medical sessions, and / or other types of multiparametric monitoring), through text extraction using speech-to-text technologies, where such information is saved in a so- called bucket associated with an identification number.

[0117] In module 2a, the data source is harmonized — for example, a curriculum vitae or a medical record in .pdf format is converted into text, or video interviews, audio recordings of medical sessions, or other types of multiparametric monitoring are transformed into text.

[0118] In module 3a, a so-called DLA (“Document Layout Analysis”) may be applied to identify sections within the file — for example, to distinguish work experience from educational background in a curriculum vitae, or, in the case of application in the healthcare sector, to identify direct factors related to the pathology as distinct from external environmental factors, such as pollution and / or nutrition. This serves to pre-process the data source in a way that best supports the questions posed to the algorithm.

[0119] In module 4a, the data are extracted — for example, in the case of video interviews, text is extracted from the video — and transformed into the appropriate format required by the artificial intelligence algorithms. This module can also perform any necessary translation of the data source, such as translating a curriculum written in Polish into English. The OCR (“Optical Character Recognition”) process enables the detection of characters contained in a document and their conversion into machine-readable digital text. A "training" phase for the OCR algorithms is also foreseen to improve performance. During this phase, the system is provided with sample images and their corresponding text in ASCII format (or similar), allowing the algorithms to calibrate on the type of text they will typically analyze. The same type of training is used to ensure the algorithms can recognize contours and reconstruct not only the text but also the page formatting.

[0120] In module 5a, the data is stored in the repository.

[0121] In module 6a, templates — either specifically designed ad hoc or imported through external libraries — are used to facilitate the so-called “parsing” process of the data source. That is, they assist in identifying the various sections within the data source, assigning priorities to them, and classifying them by type. This classification and prioritization are then used by the artificial intelligence algorithms to more effectively analyze the information and respond to specific queries.

[0122] With regard to macromodule B of processing, module lb is the core part of the artificial intelligence process. Indeed, the input data from module la, previously used and processed by modules 2a, 3a, and 4a, is here processed by various artificial intelligence algorithms — as many as there are questions for which answers are sought — generating entirely new data as output. Thanks to this, for example, a candidate or employee may be classified as a "project manager" even if the term "project manager" never appears in the data sources. Or, in the case of application in the healthcare sector, a patient may be classified according to priority or severity scales, in addition to pathology and area of intervention. Thus, it is also possible to identify potential skills (or medical conditions) with an associated proximity percentage.

[0123] In module 2b, the process carried out by module lb may be complemented — either additionally or as a substitute — by equivalent processes performed by third parties, such as “OpenAI”, in order to reduce the risk of bias and errors. By increasing both the data sources in module la and the analytical technologies in module 2b, the overall risk of system error is reduced. In module 3b, the process carried out by module lb, complemented by third parties in module 2b, may also be enriched by other "controlled" sources, such as data from software that monitors attendance, payroll, assignments, staff scheduling, and performance. In the case of application in the healthcare sector, the "controlled" sources may consist of data from medical multiparametric monitoring systems, such as blood tests, glucose monitoring, blood pressure, and heart rate. These sources are referred to as controlled because they are based on objective data — such as, in the medical context, the results of diagnostic tests.

[0124] In module 4b, the process involves the automated importation of data via APIs (Application Programming Interfaces).

[0125] In module 5b, the process involves the possible compilation of "forms" — that is, ad hoc sections of the system interface — by operators or by the candidate themselves. For example, a section where the candidate may express their job preferences or declare their skills, or, in the medical field, a section where the patient may express their dietary preferences.

[0126] In module 6b, the data is in output. These are entirely new data that classify the previously processed data on multiple levels. For example, to identify professional qualifications, the qualification "lawyer" is first identified, followed by the specialization "criminal lawyer", even when such terms are not explicitly mentioned. Similarly, in the healthcare field, to define a specific pathology — such as "hypertension" — the broader reference area, such as "cardiovascular system diseases", is first determined. The output and classification occur on multiple levels, a number of times corresponding to the thresholds set during the training phase of the algorithms, based on the proximity percentage between the new (cognitive) data and the analyzed input data.

[0127] In module 7b, the process involves reclassifying the newly generated data to associate it with the original data owner — for example, a candidate, employee, or patient — even if the data originates from multiple sources, microservices, processes, or different time periods. The reclassification may be performed by manually setting a unique identifier, such as an email address, tax code, or URL, or by using a metadata element, such as a first and last name.

[0128] In module 8b, the process generates entirely new information, such as a summary of the candidate, employee, or patient, and their “properties,” for example, hard and soft skills, or, in the medical field, allergies.

[0129] In module 9b, the process extracts all keywords and queries (i.e., sets of keywords), processes them — for example, by reducing them to their root and base form — generates a list of these, and groups or rather classifies them by areas of competence and other classifications, as in modules 6b and 10b. In module 10b, the process generates entirely new information, including: area of expertise, company size based on experience — classified, for example, into SMEs, medium-sized, and large companies — and other relevant classified information, such as the predominant sectors in which the experience was acquired, extracted from the names of companies listed among the keywords. In the healthcare domain, the entirely new information may include: environmental risk area, severity classification of the pathology, prescribed and administered drugs along with their pharmaceutical companies, and treatments received.

[0130] In module 11b, the process identifies the proximity percentage, or match, between the entirely new information, the information contained in the neural network — as in modules 3c and 4c — the information in the initial semantic knowledge graph, as in module 8c, and the “enriched” knowledge graph, as in modules 12b and 9c. Percentage thresholds are then set to display only the most relevant information, ranked by proximity.

[0131] In module 12b, the process displays on screen — either in static textual form and / or dynamic animated graphical form — the entirely new information derived from the enriched semantic knowledge graph, along with the related details. For example, when a professional qualification is displayed, it is accompanied by a descriptive textual component that specifies the "semantic meaning" of the data. Similarly, in the healthcare domain, a pathology is associated with a descriptive textual component that specifies the "semantic meaning" of the data.

[0132] In module 13b, additional information is associated with the information from module 12b, such as implication and correlation data between the entirely new data and their corresponding connection directions. For example: qualification or pathology A implies professional area or medical area X; qualification or pathology A correlates with qualification or pathology B; and qualification A implies organizational box Y. This process reveals information about potentiality — such as skills, correlations, or side effects — and their relative importance.

[0133] In module 14b, the data undergoes a merging process. For example, the “work experience” section of a curriculum vitae is merged with the “work experience” section from a professional social profde and unified into a single “work experience” section for the specific candidate or employee profile under analysis, eliminating data overlaps — such as identical information reported in the same way across multiple sources. The merging is performed in relation to the entirely new data produced by the process, including the classifications and results. The same applies in the healthcare context, where the same data — such as blood pressure readings — is merged even if collected from different sources, for example via a wearable device and / or another instrument.

[0134] With regard to macromodule (C) of training, module 1c contains the sources used for training the neural networks, as also shown in modules 2c, 3c, and 4c. In module 2c, the phase of the process dedicated to training — i.e., learning — of the neural networks is carried out. The learning may be supervised, unsupervised, or a combination of both. Supervised learning is a machine learning technique aimed at training a computer system to automatically generate output predictions based on given input, using a series of ideal examples consisting of input-output pairs provided at the outset. Unsupervised learning, on the other hand, is a machine learning technique in which the system is given a set of inputs — its experience — which it will reclassify and organize based on common features, in an attempt to reason about and predict future inputs. Unlike supervised learning, in unsupervised learning only unlabelled examples are provided, since the classes are not known a priori but must be learned automatically.

[0135] In modules 3c and 4c, the output of the process described in module 2c generates the processes of modules 3c and 4c. Specifically, module 3c includes the “basic” neural networks, which are common to all users of the platform. Module 4c, instead, includes the “custom” neural networks, which are personalized for a specific user — for example, a particular company — or for a class of users, such as a specific department or sector. To train the neural networks in module 4c, specific data from module 1c is used. For example, if the neural network is personalized for a specific company, it is that company that provides the training data, such as the texts of job descriptions, to obtain a better match between the candidate and the job offer, or, in the medical field, data from scientific studies conducted on a drug to generate more accurate predictions of possible side effects.

[0136] In module 5c, there are the basic models, classifiers, nomenclatures, databases, etc., for all the neural networks in modules 3c and 4c. These are also used as sources for training the models.

[0137] In module 6c, the basic dictionaries for all the neural networks in modules 3c and 4c are present. These can likewise be used as sources for model training. The dictionaries are used, for example, to identify keywords (module 9b), properties (module 8b), and to support all other areas of the processes described in macromodule B — for example, to identify the primary professional qualification.

[0138] Indeed, the results of the data processing lists from modules lb and 2b, which identify a list of professions or pathologies with the corresponding proximity percentage (module 11b), can be matched with the same concept as applied to keywords and queries, in order to generate a median across both and thus obtain more accurate results.

[0139] In module 7c, the semantic knowledge graphs from module 8c are present. These can also be used as sources for training the models. Similar to the dictionaries of module 6c, the "knowledge graph" is used both to generate results derived solely from module 8c and to support all other areas of the processes described in macromodule B — for example, to identify the primary professional qualification or the main pathology. As above, the results of the data processing lists from modules lb and 2b, which identify a list of professions with the corresponding proximity percentage (module 1 lb), can be matched with the same concept applied to the knowledge graph, to generate a median across both and thus achieve more accurate results.

[0140] Module 8c contains the graphs in which classifications are located — for example, those of the foundation model (module 5c) or the sources used for training the algorithms (module 1c). These graphs allow for the indication of implications and correlations — for instance, between one professional qualification and another — and the respective directions of connection, such as: qualification A implies qualification X, or pathology A implies pathology X, which in turn correlates with area Z. The graphs are used to generate entirely new data for modules 12a, 12b, and 9c. Module 8c is to be considered as a “source,” which is then processed in module 7c. The data sources in module 8c may originate from the system itself, from client companies (e.g., organizational charts or team-based research outputs, used for customizing neural networks), or from other databases or external sources. The graph is also a type of entirely new data generated by the data processing and is displayed in animated graphic form on the system's platform interface.

[0141] In module 9c, the enrichment phase is where the neural networks — either basic or customized — are enriched with information from the foundation model (module 5c), the dictionaries (module 6c), and the knowledge graphs (module 7c). The enriched information is then fed into the data processing workflow of modules lb, 2b, and 3b, and vice versa — the information from modules lb, 2b, and 3b is sent to the enrichment process. This bidirectional exchange of data allows: an exponential increase in the amount of entirely new data produced by the data processing workflow; a progressive improvement in the "intelligence" of the neural networks in modules 3c and 4c, by enriching the input information of module 1c used for training the models in module 2c.

[0142] With regard to macromodule (D) for indexing and querying the data processed by macromodule (B), module Id includes the set of input data (modules la, 1c, 8c, 4b, and 5b) and the entirely new data generated by the processing workflow (modules lb, 2b, and 3b), which are merged (module 14b) into further entirely new data. This constitutes the final result of the entire system process.

[0143] In module 2d, the data from module Id is indexed into one or more indexes. For example, in the human resources context, the indexes may include: • A. candidates for a specific job offer;

[0144] • B. candidates across the entire company;

[0145] • C. candidates processed by a specific recruiter.

[0146] Likewise, for example in the healthcare domain, the indexes may include:

[0147] • A. patients with pathologies in a specific area, such as haematology;

[0148] • B. patients followed by a specific medical team;

[0149] • C. patients followed by a specific doctor.

[0150] In module 3d, the data from module Id, indexed in module 2d, is assigned a ranking for each index, search, and query (module 5d) .

[0151] In module 4d, the input queries — or goals or questions — are provided by the user either during the pre-processing phase (modules lb, 2b, 3b), for example the parameters of a job offer or of a pathology, or during the post-data processing phase, for instance when searching within the candidate or employee database in order to "filter" the results.

[0152] In module 5d, the process of handling the queries from module 4d takes place. In module 6d, the output of the queries from module 4d is present, which filters the data from modules Id, 2d, and 3d.

[0153] In module 7d, a manual action is allowed, which the system user can perform to enrich, correct, and / or further modify the data generated by the processing in modules lb, 2b, and 3b. For example, a candidate may choose to analyze their resume to obtain proximity percentages with respect to specific areas of expertise or organizational boxes within a company chart. To do this, the user initiates the process in module la, receives a result from module Id, filters it using a query from module 4d, and views it via module 6d. At that point, the user reviews and analyzes the output. If the user decides to change the description or summary because it is not entirely accurate (data editing in module 7d), the process then generates an additional entirely new data point that is returned to the process in module 14b (“data merge”), which in turn will generate a new output in module Id and new indexes and rankings in modules 2d and 3d. Similarly, in the healthcare context, the patient can perform the same manual action, where the "areas of expertise" are interpreted as "pathologies." The manual action may also be a simple one — for instance, clicking a button to hire a candidate implicitly communicates that the result, and thus the data, is appropriate to the input query from module 4d.

[0154] In module 8d, the “processed” data, as described in module 7d, can be used as an additional source (module 1c) for further training (module 2c) of the neural networks (modules 3c and 4c), which will then generate further entirely new data. This process defines the system's neural networks as "self-learning". - l -

[0155] According to a preferred embodiment, the system according to the invention may be integrated into a multimedia totem configured to collect interviews in video and / or audio format. Such totem is suitable for allowing personnel, either already employed by the company or newly onboarded, to record video or audio content, for instance in response to specific questions. The totem is advantageously portable and can be deployed across the various production units of the company. Advantageously, such totem is configured to operate as an interface for collecting at least some of the data referred to in phase (la) of data collection. The totem is also configured to carry out phase (2a) of pre-processing and converting said data into text and phase (4a) of transforming the data according to the method of the invention.

[0156] It has been found in practice that the method and system for multiparametric processing of information from multidimensional and / or multitemporal datasets, according to the present invention, fulfil the intended objectives and purposes, as they allow original data to be transformed — through techniques of data processing, synthesis, and classification — into new datasets.

[0157] Another advantage of the method and system according to the invention is that the original data can be grouped into datasets based on the principle of proximity, associating them with a specific reference category, and combining them in a graph-based database that uses nodes and edges to represent and store multidimensional and multitemporal information, display correlations, implications, distances, and inferences, define the direction of links among data, datasets, and classifiers — including node membership in the graph and identification of semantic relationships, known as "Relationship Extraction" — thus generating new datasets.

[0158] A further advantage of the method and system according to the invention is the ability to provide entirely new data by exploiting semantic relationships among the original datasets, the datasets generated using the method, the classifiers, and both public and private knowledge graphs, also by associating the data with the original classifier match.

[0159] Yet another advantage of the system according to the invention is that the new data is enriched with descriptive, codified information that enables proper interpretation.

[0160] An additional advantage of the system according to the invention lies in the informational enrichment of the original data, its reprocessing, and its redistribution, thereby enhancing information that is not immediately perceivable or analyzable.

[0161] Another advantage of the method and system according to the invention is its adaptability to all contexts in which it is necessary to handle multiparametric, multidimensional, and / or multitemporal data, even when such data originate simultaneously from multiple sources and in different formats. A further advantage of the method and system according to the invention is its particular applicability in the field of human resources, in the analysis of medical records, and in all areas where current and historical data must be analyzed to understand their classification (or correspondence) and / or their future meaning (or probability).

[0162] The proposed invention differentiates itself from traditional OCR systems through its ability to process heterogeneous multimedia data, including text, images, videos and audio recordings, using an unsupervised artificial intelligence approach. This enables a significant reduction in computational resources required, thanks to techniques such as inference skipping, intelligent batch processing, and model quantization or pruning. These mechanisms prevent redundant processing of repetitive or low-informative content, optimizing the use of CPU, GPU, and RAM. Furthermore, the automatic extraction of text from videos (e.g., video interviews) and images using advanced OCR algorithms allows the transformation of unstructured data into structured semantic information, enhancing the quality and relevance of the generated output.

[0163] Unlike traditional OCR systems, which process each document independently and do not improve with increased data volume, the described system leverages unsupervised learning to recognize patterns, inferences, and correlations in multimedia data. As the system processes an increasing number of documents, it refines its internal models, enhancing the coherence of semantic clusters, the quality of embeddings, and the ability to identify complex structures within the data. This approach enables high scalability, with progressive improvement in analytical performance as the dataset grows, at the same time facilitating the identification of insights and hidden relationships within multimedia data.

[0164] The adoption of optimization techniques such as inference skipping, intelligent batch processing, and model quantization enables the system to significantly reduce processing times and energy consumption, especially when analyzing large volumes of multimedia data. Unlike traditional OCR systems, which require linear and uniform processing for each document, the proposed system dynamically adapts hardware resource usage based on the complexity and content of the input data, enhancing overall efficiency and reducing operational costs.

[0165] Another advantage of the method and system according to the invention lies in the ability to accelerate decision-making processes in which the method is implemented, by reducing the amount of manual work required by operators. For example, in the HR sector, the method and system according to the invention allow for the fast and reliable generation of a shortlist of candidates for a given position, based on the published job offer, any additional requirements, and the resumes uploaded by the candidates. Furthermore, it is possible to standardize processes by applying rules and filters in a consistent manner, as well as ensure traceability and easy consultation of the processed data.

[0166] The method and system, as conceived, may be subject to numerous modifications and variations, all falling within the scope of the inventive concept. Furthermore, all details may be replaced with technically equivalent elements.

Claims

CLAIMS1. A method for multiparametric processing of information from multidimensional and / or multitemporal datasets comprising the following four macrostages: a) macrostage (A) of input and pre-processing of data relating to an entity; b) macrostage (B) of processing said data by means of a neural network-based artificial intelligence algorithm; c) macrostage (C) of training said neural network; d) macrostage (D) of indexing and querying said data processed by said macrostage (B) of processing; characterized in that said macrostage (A) of input and pre-processing of data comprises at least the following phases: phase (la) of collecting data relating to an entity, in which a plurality of data from one or more data sources is collected, said data comprising at least text documents, images, videos and audio recordings; phase (2a) of pre-processing and converting said data into text, in which said text documents, images, videos and audio recordings are converted into textual formats; phase (4a) of transforming said data in which said data are extracted and transformed in a format adapted to be processed by said neural network-based artificial intelligence algorithm; phase (5a) of storing said data, in which said data are stored in a repository; said macrostage (C) of training said neural network comprises at least the following phases: phase (1c) of collecting training data for training said neural network, in which training data are collected from a plurality of sources; phase (2c) of training said neural network, in which supervised and / or unsupervised learning is performed to train said neural network; phase of generating said neural network comprising a phase (4c) of generating a customized neural network, in which a customized neural network is implemented for a specific user and / or for a specific class of users; said macrostage (B) of processing comprises at least the following phases: phase (lb) of processing said pre-processed data in said macrostage (A) of input and pre-processing of data, in which said pre-processed data are processed by means of a neural network-based artificial intelligence algorithm based on said neural network generated in said phase (4c) of generating said neural network and in which new data associated with said pre-processed data are generated;phase (14b) of merging said data, in which corresponding data from different sources are merged; said macrostage (D) of indexing and querying comprises the following phases: phase (Id) of merging said data relating to said entity and said new data, in which input data relating to said entity are merged with said new data generated in said macrostage (B) of processing and in which merged data are generated and stored in said repository; phase (2d) of indexing the merged data, in which said merged data are indexed according to one or more indexes and indexed data are generated; phase (4d) of collecting queries from a user, in which the insertion of one or more search queries by said user is allowed; phase (5d) of processing said one or more queries, in which said one or more search queries are processed; phase (6d) of outputting said one or more queries, in which the results of said one or more search queries are filtered and visualized.

2. The method according to the preceding claim, wherein said macrostage (A) of input and pre-processing of data comprises one or more of: a phase (3a) of analyzing said data when collected in the form of a text document, in which a document layout analysis (DLA) is applied to identify and separate different sections present in said document; in said phase (4a) of transforming said data, translating said data into a selected language; a phase (6a) of applying templates for parsing analysis, in which one or more templates are applied to said data to facilitate parsing of said data by classifying and prioritizing identified sections in said data, obtaining pre-processed data.

3. The method according to one or more of the preceding claims, wherein said macrostage (B) of processing comprises one or more of: a phase (2b) of parallel control, in which an analysis with third-party artificial intelligence algorithms is performed and in which additional information from said third-party artificial intelligence algorithms is incorporated into said data; a phase (3b) of controlled integration of said data, in which additional information from controlled data sources is incorporated into said data; a phase (4b) of automated importation of said additional information, in which said additional information is automatically imported into said data through anAPI (Application Programming Interface); a phase (5b) of manual compilation of forms, in which the manual insertion of information to integrate said data is allowed; a phase (6b) of classifying said new data, in which new classifications of said new data are generated by classifying them at multiple levels; a phase (7b) of reclassifying said new data, in which said new data are reclassified to associate them with said entity, using unique identifiers of said entity; a phase (8b) of generating summaries and properties, in which summaries of said new data are generated and in which descriptive elements of the properties of said entity are generated; a phase (9b) of extracting keywords, in which keywords and / or groups of keywords are extracted and classified from said new data; a phase (10b) of generating new information, in which new information is generated based on said new data; a phase (11b) of generating a proximity percentage, in which the proximity percentage between said new data and the information present in said phase (4c) of generating said neural network and / or in said phase (8c) of collecting sources for the preparation of said knowledge semantic graphs of said phase (7c) of preparing knowledge semantic graphs and / or in said phase (9c) of enriching said neural network is calculated; a phase (12b) of visualizing said knowledge semantic graphs, in which graphic and textual outputs of said knowledge semantic graphs applied to said new data are generated; a phase (13b) of identifying implications and correlations, in which implications and correlations between said new data are identified.

4. The method according to one or more of the preceding claims, wherein said macrostage (C) of training said neural network comprises one or more of: a phase (5c) of preparing basic knowledge models, in which basic knowledge models are prepared to be used as additional sources of training data for the generation of said neural network in said phase (4c) of generating a customized neural network; a phase (6c) of preparing basic dictionaries, in which dictionaries are prepared to be used as additional sources of training data for the generation of said neural network in said phase (4c) of generating a customized neural network;a phase (7c) of preparing knowledge semantic graphs, in which semantic graphs are prepared to be used as additional sources of training data for the generation of said neural network in said phase (4c) of generating a customized neural network; a phase (8c) of collecting sources for the preparation of said knowledge semantic graphs of said phase (7c) of preparing knowledge semantic graphs, in which implications and correlations between the elements of said knowledge semantic graphs are generated; a phase (9c) of enriching said neural network, in which said neural network generated in said phase (4c) of generating said neural network is enriched with information generated at least in said phase (lb) of processing said pre-processed data and in one or more of: said phase (5c) of preparing basic knowledge models, said phase (6c) of preparing basic dictionaries, and said phase (7c) of preparing knowledge semantic graphs;5. The method according to the preceding claim, wherein in said phase (9c) of enriching said neural network, said neural network generated in said phase (4c) of generating said neural network is enriched with information originating not only from said phase (lb) of processing said pre- processed data but also from said phase (2b) of parallel control and / or from said phase (3b) of controlled integration of said data.

6. The method according to one or more of the preceding claims, wherein said macrostage (D) of indexing and querying comprises one or more of: a phase (3d) of ranking said indexed data, in which a ranking for each of said one or more indexes is assigned to said indexed data; a phase (7d) of manually manipulating said new data generated in said macrostage (B) of processing, in which said user is allowed to enrich, correct, and modify said new data generated in said macrostage (B) of processing; a phase (8d) of reusing said manipulated new data in said phase (7d) of manual manipulation, in which said manipulated new data is used as an additional source of training data for the generation of said neural network in said phase (4c) of generating a customized neural network and in which further training cycles of said neural network are initiated based on said phase (2c) of training said neural network.

7. A system for multiparametric processing of information from multidimensional and / or multitemporal datasets comprising means for executing the method according to the claim 1, said means comprising the following four macrostages:a) a macrostage (A) of input and pre-processing of data related to an entity; b) a macrostage (B) of processing said data through an artificial intelligence algorithm based on a neural network; c) a macrostage (C) of training said neural network; d) a macrostage (D) of indexing and querying said data processed by said macrostage (B) of processing; characterized in that said macrostage (A) of input and pre-processing of data comprises at least the following modules: module (la) of collecting data related to an entity, configured to collect a plurality of data from one or more data sources, said data comprising at least textual documents, images, videos and audio recordings; module (2a) of pre-processing and converting said data into text, in which said textual documents, images, videos and audio recordings are converted into textual formats; module (4a) of transforming said data in which said data are extracted and transformed in a format adapted to be processed by said neural network-based artificial intelligence algorithm; module (5a) of storing said data, configured to store said data in a repository; said macrostage (C) of training said neural network comprises at least the following modules: module (1c) of collecting training data for training said neural network, configured to collect training data from a plurality of sources; module (2c) of training said neural network, configured to perform supervised and / or unsupervised learning to train said neural network; module of generating said neural network comprising a module (4c) of generating a personalized neural network, configured to implement a personalized neural network for a specific user and / or for a class of specific users; said macrostage (B) of processing comprises at least the following modules: module (lb) of processing said pre-processed data from said macrostage (A) of input and pre-processing of data, configured to process said pre-processed data through an artificial intelligence algorithm based on said neural network generated in said module (4c) of generating said neural network and to generate new data associated with said pre-processed data;module (14b) of merging said data, configured to merge corresponding data from different sources; said macrostage (D) of indexing and querying comprises the following modules: module (Id) of merging said data related to said entity and said new data, configured to merge the input data related to said entity with said new data generated by said macrostage (B) of processing and to generate merged data, stored in said repository; module (2d) of indexing the merged data, configured to index said merged data according to one or more indices and to generate indexed data; module (4d) of collecting queries from a user, configured to allow the insertion of one or more search queries by said user; module (5d) of processing said one or more queries, configured to process said one or more search queries; module (6d) of outputting said one or more queries, configured to filter and display the results of said one or more search queries.

8. The system according to the preceding claim, wherein said macrostage (A) of input and pre-processing of data comprises one or more of: a module (3a) of analyzing said data when collected in the form of a text document, configured to apply document layout analysis (DLA) to identify and separate different sections present in said document; said module (4a) of transforming said data, being configured to translate said data into a selected language; a module (6a) of applying templates for parsing analysis, configured to apply one or more templates to said data to facilitate the parsing of said data, classifying and prioritizing sections identified in said data, obtaining pre-processed data.

9. The system according to claim 7 or 8, wherein said macrostage (B) of processing comprises one or more of: a module (2b) of parallel control, configured to integrate the analysis with artificial intelligence algorithms of third parties and to incorporate into said data additional information from said third-party artificial intelligence algorithms; module (3b) of controlled integration of said data, configured to incorporate into said data additional information from controlled data sources;a module (4b) of automated importation of said additional information, configured to automatically import said additional information into said data through an API (Application Programming Interface); a module (5b) of manual compilation of forms, configured to allow the manual insertion of information to integrate said data; module (6b) of classifying said new data, configured to generate new classifications of said new data by classifying them at multiple levels; a module (7b) of reclassifying said new data, configured to reclassify said new data to associate them with said entity, using unique identifiers of said entity; a module (8b) of generating summaries and properties, configured to generate summaries of said new data and to generate descriptive elements of the properties of said entity; a module (9b) of extracting keywords, configured to extract and classify keywords and / or groups of keywords from said new data; a module (10b) of generating new information, configured to generate new information based on said new data; a module (11b) of generating a proximity percentage, configured to calculate the proximity percentage between said new data and the information contained in said module (4c) of generating said neural network and / or in said module (8c) of collecting sources for preparing said semantic knowledge graphs of said module (7c) of preparing semantic knowledge graphs and / or in said module (9c) of enriching said neural network; a module (12b) of visualizing said semantic knowledge graphs, configured to generate graphical and textual outputs of said semantic knowledge graphs applied to said new data; a module (13b) of identifying implications and correlations, configured to identify implications and correlations between said new data.

10. The system according to one or more of claims 7 to 9, wherein said macrostage (C) of training said neural network comprises one or more of: a module (5c) of preparing basic knowledge models, configured to prepare basic knowledge models, usable as additional sources of training data for generating said neural network in said module (4c) of generating a personalized neural network;a module (6c) of preparing basic dictionaries, configured to prepare dictionaries usable as additional sources of training data for generating said neural network in said module (4c) of generating a personalized neural network; a module (7c) of preparing semantic knowledge graphs, configured to prepare semantic graphs usable as additional sources of training data for generating said neural network in said module (4c) of generating a personalized neural network; a module (8c) of collecting sources for preparing said semantic knowledge graphs of said module (7c) of preparing semantic knowledge graphs, configured to generate implications and correlations between the elements of said semantic knowledge graphs; a module (9c) of enriching said neural network, configured to enrich said neural network generated by said module (4c) of generating said neural network with information originating at least from said module (lb) of processing said pre- processed data and from one or more of: said module (5c) of preparing basic knowledge models, said module (6c) of preparing basic dictionaries, and said module (7c) of preparing semantic knowledge graphs.

11. The system according to the preceding claim, wherein said module (9c) of enriching said neural network is configured to enrich said neural network generated by said module (4c) of generating said neural network with information originating not only from said module (lb) of processing said pre-processed data but also from said module (2b) of parallel control and / or from said module (3b) of controlled integration of said data.

12. The system according to one or more of claims 7 to 11, wherein said macrostage (D) of indexing and querying comprises one or more of: module (3d) of ranking said indexed data, configured to assign a ranking for each of said one or more indices to said indexed data; module (7d) of manually manipulating said new data generated by said macrostage (B) of processing, configured to allow said user to enrich, correct and modify said new data generated by said macrostage (B) of processing; module (8d) of reusing said new data manipulated in said module (7d) of manual manipulation, configured to use said manipulated new data as an additional source of training data for generating said neural network in said module (4c) of generating a personalized neural network and to initiate further training cycles of said neural network through said module (2c) of training said neural network.

13. A computer program comprising instructions which, when the program is executed by a computer, causes said computer to execute the method according to claim 1.

Images