Claims:

1. A processor implemented method (200), the method comprising:
receiving, via one or more hardware processors, an incident report, wherein the incident report comprises a plurality of words (202);
computing, via the one or more hardware processors, a plurality of features corresponding to each of the plurality of words associated with the incident report using a pretrained Bidirectional Encoder Representations from Transformers (BERT) model (204);
computing, via the one or more hardware processors, a plurality of reduced features based on the plurality of features using a Fully Connected Neural Network (FCNN) (206);
obtaining, via the one or more hardware processors, a named entity information for each of the plurality of words using a Natural Language Processing (NLP) tool (208);
obtaining, via the one or more hardware processors, a Parts Of Speech (POS) information for each of the plurality of words using the NLP tool (210);
obtaining, via the one or more hardware processors, a plurality of feature vectors corresponding to each of the plurality of words by concatenating the plurality of reduced features, the named entity information and the POS information (212);
computing, via the one or more hardware processors, a plurality of contextual feature values corresponding to each of the plurality of words based on the plurality of feature vectors using a pretrained Bidirectional Long Short Term Memory (BiLSTM), wherein the BiLSTM is pretrained using transfer learning method (214);
annotating, via the one or more hardware processors, each of the plurality of words with a corresponding label based on the plurality of contextual feature values using a pretrained Conditional Random Fields (CRF) model (216);
identifying, via the one or more hardware processors, a plurality of events from the annotated plurality of words based on the corresponding label (218); and
validating, via the one or more hardware processors, the identified plurality of events based on the plurality of contextual feature values using a clustering based validation technique (220).
2. The method as claimed in claim 1, wherein step of validating the identified plurality of events based on the plurality of contextual feature values using the clustering based validation technique comprises:
receiving the plurality of contextual feature values from the pretrained BiLSTM and a label corresponding to each of the plurality of contextual feature values from the CRF model;
generating a plurality of feature clusters by clustering each of the plurality of contextual feature values using a complete agglomerative clustering;
for each feature cluster, comparing each label corresponding to each of the plurality of contextual feature values with a plurality of standard event labels;
setting a flag corresponding to each feature cluster to one if the label corresponding to each of the plurality of contextual feature values are similar; and
setting a validation flag to one if the flag corresponding to each of the plurality of feature clusters is set to one, wherein the event identification is considered as valid if the validation flag is set to one.
3. The method as claimed in claim 1, wherein the method of pretraining the BiLSTM using transfer learning comprises:
training the BiLSTM based on a general purpose annotated dataset by:
receiving a general-purpose annotated dataset comprising a first set of annotated words;
computing a first set of features corresponding to each word from the first set of annotated words associated with the general-purpose annotated dataset using a BERT model;
computing a first set of reduced features based on the first set of features using the Fully Connected Neural Network (FCNN); and
training the BiLSTM based on the first set of reduced features and the first set of annotated words; and
finetuning the BiLSTM based on a small annotated industrial incident dataset by:
receiving an annotated industrial incident dataset comprising a second set of annotated words;
computing a second set of features corresponding to each word from of the second set of annotated words using the BERT model;
computing a second set of reduced features based on the second set of features using the Fully Connected Neural Network (FCNN); and
training the BiLSTM based on the second set of reduced features and the second set of annotated words.
4. A system (100) comprising:
at least one memory (104) storing programmed instructions; one or more Input /Output (I/O) interfaces (112); and one or more hardware processors (102) operatively coupled to the at least one memory (104), wherein the one or more hardware processors (102) are configured by the programmed instructions to:
receive an incident report, wherein the incident report comprises a plurality of words;
compute a plurality of features corresponding to each of the plurality of words associated with the incident report using a pretrained Bidirectional Encoder Representations from Transformers (BERT) model;
compute a plurality of reduced features based on the plurality of features using a Fully Connected Neural Network (FCNN);
obtain a named entity information for each of the plurality of words using a Natural Language Processing (NLP) tool;
obtain a Parts Of Speech (POS) information for each of the plurality of words using the NLP tool;
obtain a plurality of feature vectors corresponding to each of the plurality of words by concatenating the plurality of reduced features, the named entity information and the POS information;
compute a plurality of contextual feature values corresponding to each of the plurality of words based on the plurality of feature vectors using a pretrained Bidirectional Long Short Term Memory (BiLSTM), wherein the BiLSTM is pretrained using transfer learning method;
annotate each of the plurality of words with a corresponding label based on the plurality of contextual feature values using a pretrained Conditional Random Fields (CRF) model;
identify a plurality of events from the annotated plurality of words based on the corresponding label; and
validate the identified plurality of events based on the plurality of contextual feature values using a clustering based validation technique.

5. The system of claim 4, wherein step of validating the identified plurality of events based on the plurality of contextual feature values using the clustering based validation technique comprises:
receiving the plurality of contextual feature values from the pretrained BiLSTM and a label corresponding to each of the plurality of contextual feature values from the CRF model;
generating a plurality of feature clusters by clustering each of the plurality of contextual feature values using a complete agglomerative clustering;
for each feature cluster, comparing each label corresponding to each of the plurality of contextual feature values with a plurality of standard event labels;
setting a flag corresponding to each feature cluster to one if the label corresponding to each of the plurality of contextual feature values are similar; and
setting a validation flag to one if the flag corresponding to each of the plurality of feature clusters is set to one, wherein the event identification is considered as valid if the validation flag is set to one.
6. The system of claim 4, wherein the method of pretraining the BiLSTM using transfer learning comprises:
training the BiLSTM based on a general purpose annotated dataset by:
receiving a general-purpose annotated dataset comprising a first set of annotated words;
computing a first set of features corresponding to each word from the first set of annotated words associated with the general-purpose annotated dataset using a BERT model;
computing a first set of reduced features based on the first set of features using the Fully Connected Neural Network (FCNN); and
training the BiLSTM based on the first set of reduced features and the first set of annotated words; and
finetuning the BiLSTM based on a small annotated industrial incident dataset by:
receiving an annotated industrial incident dataset comprising a second set of annotated words;
computing a second set of features corresponding to each word from of the second set of annotated words using the BERT model;
computing a second set of reduced features based on the second set of features using the Fully Connected Neural Network (FCNN); and
training the BiLSTM based on the second set of reduced features and the second set of annotated words.
, Description:FORM 2

THE PATENTS ACT, 1970
(39 of 1970)
&
THE PATENT RULES, 2003

COMPLETE SPECIFICATION
(See Section 10 and Rule 13)

Title of invention:
METHOD AND SYSTEM FOR TRANSFER LEARNING BASED EVENT IDENTIFICATION AND VALIDATION

Applicant:
Tata Consultancy Services Limited
A company Incorporated in India under the Companies Act, 1956
Having address:
Nirmal Building, 9th Floor,
Nariman Point, Mumbai 400021,
Maharashtra, India

The following specification particularly describes the invention and the manner in which it is to be performed.
TECHNICAL FIELD
[001] The disclosure herein generally relates to the field of machine learning and, more particular, to a method and system for transfer learning based event identification and validation.
BACKGROUND
[002] The industrial revolution has had a profound effect on the socio-political fabric of the world. Economic progress of societies has been highly correlated with their degree of industrialization. However, one of the flip sides of this progress has been the cost of large industrial accidents in terms of injuries to workers, damage to material and property as well as the irreparable loss of innocent human lives. Hence event reporting and analysis has become a key step in industrial process. Events are specific occurrences that appear in the text to denote happenings or changes in states of the involved participants. Extracting events from reports on incidents is an important functionality in industries for ensuring safety of workers and the industry itself.
[003] Most of the conventional methods for incident report analysis are manual in nature. There is very less focus on automated processing of incident reports. Some of the automated event identification systems currently available are implemented using deep learning approaches and rule-based methods. The deep learning methods miss out many important verb-based events while the rule-based approach fails to identify nominal events correctly as it doesn’t observe the context of a noun while deciding its event nature. Hence there is a challenge in developing a system for extracting verb-based event predictions along with nominal event predictions in incidental reports.

SUMMARY
[004] Embodiments of the present disclosure present technological improvements as solutions to one or more of the above-mentioned technical problems recognized by the inventors in conventional systems. For example, in one embodiment, a method for transfer learning-based event identification and validation is provided. The method includes receiving, by one or more hardware processors, an incident report, wherein the incident report comprises a plurality of words. Further, the method includes computing by the one or more hardware processors, a plurality of features corresponding to each of the plurality of words associated with the incident report using a pretrained Bidirectional Encoder Representations from Transformers (BERT) model. Furthermore, the method includes computing by the one or more hardware processors, a plurality of reduced features based on the plurality of features using a Fully Connected Neural Network (FCNN). Furthermore, the method includes obtaining by the one or more hardware processors, a named entity information for each of the plurality of words using a Natural Language Processing (NLP) tool. Furthermore, the method includes obtaining by the one or more hardware processors, a Parts Of Speech (POS) information for each of the plurality of words using the NLP tool. Furthermore, the method includes obtaining by the one or more hardware processors, a plurality of feature vectors corresponding to each of the plurality of words by concatenating the plurality of reduced features, the named entity information, and the POS information. Furthermore, the method includes computing (214), via the one or more hardware processors, a plurality of contextual feature values corresponding to each of the plurality of words based on the plurality of feature vectors using a pretrained Bidirectional Long Short Term Memory (BiLSTM), wherein the BiLSTM is pretrained using transfer learning method. Furthermore, the method includes annotating by the one or more hardware processors, each of the plurality of words with a corresponding label based on the plurality of contextual feature values using a pretrained Conditional Random Fields (CRF) model. Furthermore, the method includes identifying by the one or more hardware processors, a plurality of events from the annotated plurality of words based on the corresponding label. Finally, the method includes validating by the one or more hardware processors, the identified plurality of events based on the plurality of contextual feature values using a clustering based validation technique.
[005] In another aspect, a system for transfer learning based event identification and validation is provided. The system includes at least one memory storing programmed instructions, one or more Input /Output (I/O) interfaces, and one or more hardware processors operatively coupled to the at least one memory, wherein the one or more hardware processors are configured by the programmed instructions to receive an incident report, wherein the incident report comprises a plurality of words. Further, the one or more hardware processors are configured by the programmed instructions to compute a plurality of features corresponding to each of the plurality of words associated with the incident report using a pretrained Bidirectional Encoder Representations from Transformers (BERT) model. Furthermore, the one or more hardware processors are configured by the programmed instructions to compute a plurality of reduced features based on the plurality of features using a Fully Connected Neural Network (FCNN). Furthermore, the one or more hardware processors are configured by the programmed instructions to obtain a named entity information for each of the plurality of words using a Natural Language Processing (NLP) tool. Furthermore, the one or more hardware processors are configured by the programmed instructions to obtain a Parts Of Speech (POS) information for each of the plurality of words using the NLP tool. Furthermore, the one or more hardware processors are configured by the programmed instructions to obtain a plurality of feature vectors corresponding to each of the plurality of words by concatenating the plurality of reduced features, the named entity information, and the POS information. Furthermore, the one or more hardware processors are configured by the programmed instructions to compute a plurality of contextual feature values corresponding to each of the plurality of words based on the plurality of feature vectors using a pretrained Bidirectional Long Short Term Memory (BiLSTM), wherein the BiLSTM is pretrained using transfer learning method. Furthermore, the one or more hardware processors are configured by the programmed instructions to annotate each of the plurality of words with a corresponding label based on the plurality of contextual feature values using a pretrained Conditional Random Fields (CRF) model. Furthermore, the one or more hardware processors are configured by the programmed instructions to identify a plurality of events from the annotated plurality of words based on the corresponding label. Finally, the one or more hardware processors are configured by the programmed instructions to validate the identified plurality of events based on the plurality of contextual feature values using a clustering based validation technique.
[006] In yet another aspect, a computer program product including a non-transitory computer-readable medium having embodied therein a computer program for transfer learning based event identification and validation is provided. The computer readable program, when executed on a computing device, causes the computing device to receive an incident report, wherein the incident report comprises a plurality of words. Further, the computer readable program, when executed on a computing device, causes the computing device to compute a plurality of features corresponding to each of the plurality of words associated with the incident report using a pretrained Bidirectional Encoder Representations from Transformers (BERT) model. Furthermore, the computer readable program, when executed on a computing device, causes the computing device to compute a plurality of reduced features based on the plurality of features using a Fully Connected Neural Network (FCNN). Furthermore, the computer readable program, when executed on a computing device, causes the computing device to obtain a named entity information for each of the plurality of words using a Natural Language Processing (NLP) tool. Furthermore, the computer readable program, when executed on a computing device, causes the computing device to obtain a Parts Of Speech (POS) information for each of the plurality of words using the NLP tool. Furthermore, the computer readable program, when executed on a computing device, causes the computing device to obtain a plurality of feature vectors corresponding to each of the plurality of words by concatenating the plurality of reduced features, the named entity information and the POS information. Furthermore, the computer readable program, when executed on a computing device, causes the computing device to compute a plurality of contextual feature values corresponding to each of the plurality of words based on the plurality of feature vectors using a pretrained Bidirectional Long Short Term Memory (BiLSTM), wherein the BiLSTM is pretrained using transfer learning method. Furthermore, the computer readable program, when executed on a computing device, causes the computing device to annotate each of the plurality of words with a corresponding label based on the plurality of contextual feature values using a pretrained Conditional Random Fields (CRF) model. Furthermore, the computer readable program, when executed on a computing device, causes the computing device to identify a plurality of events from the annotated plurality of words based on the corresponding label. Finally, the computer readable program, when executed on a computing device, causes the computing device to validate the identified plurality of events based on the plurality of contextual feature values using a clustering based validation technique.
[007] It is to be understood that both the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the invention, as claimed.
BRIEF DESCRIPTION OF THE DRAWINGS

[008] The accompanying drawings, which are incorporated in and constitute a part of this disclosure, illustrate exemplary embodiments and, together with the description, explain the disclosed principles:
[009] FIG. 1 is a functional block diagram of a system for transfer learning based event identification and validation, in accordance with some embodiments of the present disclosure.
[0010] FIGS. 2A and 2B are exemplary flow diagrams illustrating a method for transfer learning based event identification and validation, implemented by the system of FIG. 1, in accordance with some embodiments of the present disclosure.
[0011] FIG. 3 is an example process flow architecture for the processor implemented method for transfer learning based event identification and validation implemented by the system of FIG. 1, in accordance with some embodiments of the present disclosure.
[0012] FIG. 4 is an example process flow architecture for a clustering based validation technique associated with the processor implemented method for transfer learning based event identification and validation implemented by the system of FIG. 1, in accordance with some embodiments of the present disclosure.

DETAILED DESCRIPTION OF EMBODIMENTS
[0013] Exemplary embodiments are described with reference to the accompanying drawings. In the figures, the left-most digit(s) of a reference number identifies the figure in which the reference number first appears. Wherever convenient, the same reference numbers are used throughout the drawings to refer to the same or like parts. While examples and features of disclosed principles are described herein, modifications, adaptations, and other implementations are possible without departing from the spirit and scope of the disclosed embodiments.
[0014] Embodiments herein provide a method and system for transfer learning based event identification and validation for identifying events from any incident report. Initially, the system receives an incident report. The incident report includes a plurality of words. Further, a plurality of features corresponding to each of the plurality of words associated with the incident report are computed using a pretrained Bidirectional Encoder Representations from Transformers (BERT) model. After computing the computing plurality of features, a plurality of reduced features are computed based on the plurality of features using a Fully Connected Neural Network (FCNN). A named entity information is obtained for each of the plurality of words using a Natural Language Processing (NLP) tool. Similarly, a Parts Of Speech (POS) information for each of the plurality of words is obtained using the NLP tool. Further, a plurality of feature vectors corresponding to each of the plurality of words are generated by concatenating the plurality of reduced features, the named entity information and the POS information. After concatenating, a plurality of contextual feature values corresponding to each of the plurality of words are computed based on the plurality of feature vectors using a pretrained Bidirectional Long Short Term Memory (BiLSTM), wherein the BiLSTM is pretrained using transfer learning method. Further, each of the plurality of words are annotated with a corresponding label based on the plurality of contextual feature values using a pretrained Conditional Random Fields (CRF) model. After annotation, a plurality of events from the annotated plurality of words are identified based on the corresponding label. The annotated events are further validated based on the plurality of contextual feature values using a clustering based validation technique.
[0015] Referring now to the drawings, and more particularly to FIGS. 1 through 4, where similar reference characters denote corresponding features consistently throughout the figures, there are shown preferred embodiments and these embodiments are described in the context of the following exemplary system and/or method.
[0016] FIG. 1 is a functional block diagram of a system 100 for transfer learning based event identification and validation, according to some embodiments of the present disclosure. The system 100 includes or is otherwise in communication with hardware processors 102, at least one memory such as a memory 104, an I/O interface 112. The hardware processors 102, memory 104, and the Input /Output (I/O) interface 112 may be coupled by a system bus such as a system bus 108 or a similar mechanism. In an embodiment, the hardware processors 102 can be one or more hardware processors.
[0017] The I/O interface 112 may include a variety of software and hardware interfaces, for example, a web interface, a graphical user interface, and the like. The I/O interface 112 may include a variety of software and hardware interfaces, for example, interfaces for peripheral device(s), such as a keyboard, a mouse, an external memory, a printer, and the like. Further, the I/O interface 112 may enable the system 100 to communicate with other devices, such as web servers, and external databases.
[0018] The I/O interface 112 can facilitate multiple communications within a wide variety of networks and protocol types, including wired networks, for example, local area network (LAN), cable, etc., and wireless networks, such as Wireless LAN (WLAN), cellular, or satellite. For the purpose, the I/O interface 112 may include one or more ports for connecting several computing systems with one another or to another server computer. The I/O interface 112 may include one or more ports for connecting several devices to one another or to another server.
[0019] The one or more hardware processors 102 may be implemented as one or more microprocessors, microcomputers, microcontrollers, digital signal processors, central processing units, node machines, logic circuitries, and/or any devices that manipulate signals based on operational instructions. Among other capabilities, the one or more hardware processors 102 is configured to fetch and execute computer-readable instructions stored in the memory 104.
[0020] The memory 104 may include any computer-readable medium known in the art including, for example, volatile memory, such as static random access memory (SRAM) and dynamic random access memory (DRAM), and/or non-volatile memory, such as read only memory (ROM), erasable programmable ROM, flash memories, hard disks, optical disks, and magnetic tapes. In an embodiment, the memory 104 includes a plurality of modules 106. The memory 104 also includes a data repository (or repository) 110 for storing data processed, received, and generated by the plurality of modules 106.
[0021] The plurality of modules 106 include programs or coded instructions that supplement applications or functions performed by the system 100 for transfer learning based event identification and validation. The plurality of modules 106, amongst other things, can include routines, programs, objects, components, and data structures, which performs particular tasks or implement particular abstract data types. The plurality of modules 106 may also be used as, signal processor(s), node machine(s), logic circuitries, and/or any other device or component that manipulates signals based on operational instructions. Further, the plurality of modules 106 can be used by hardware, by computer-readable instructions executed by the one or more hardware processors 102, or by a combination thereof. The plurality of modules 106 can include various sub-modules (not shown). The plurality of modules 106 may include computer-readable instructions that supplement applications or functions performed by the system 100 for transfer learning based event identification and validation.
[0022] The data repository (or repository) 110 may include a plurality of abstracted piece of code for refinement and data that is processed, received, or generated as a result of the execution of the plurality of modules in the module(s) 106.
[0023] Although the data repository 110 is shown internal to the system 100, it will be noted that, in alternate embodiments, the data repository 110 can also be implemented external to the system 100, where the data repository 110 may be stored within a database (not shown in FIG. 1) communicatively coupled to the system 100. The data contained within such external database may be periodically updated. For example, new data may be added into the database (not shown in FIG. 1) and/or existing data may be modified and/or non-useful data may be deleted from the database (not shown in FIG. 1). In one example, the data may be stored in an external system, such as a Lightweight Directory Access Protocol (LDAP) directory and a Relational Database Management System (RDBMS).
[0024] FIGS. 2A and 2B are exemplary flow diagrams illustrating a method 200 for transfer learning based event identification and validation implemented by the system of FIG. 1 according to some embodiments of the present disclosure. In an embodiment, the system 100 includes one or more data storage devices or the memory 104 operatively coupled to the one or more hardware processor(s) 102 and is configured to store instructions for execution of steps of the method 200 by the one or more hardware processors 102. The steps of the method 200 of the present disclosure will now be explained with reference to the components or blocks of the system 100 as depicted in FIG. 1 and the steps of flow diagram as depicted in FIG. 2A and 2B. The method 200 may be described in the general context of computer executable instructions. Generally, computer executable instructions can include routines, programs, objects, components, data structures, procedures, modules, functions, etc., that perform particular functions or implement particular abstract data types. The method 200 may also be practiced in a distributed computing environment where functions are performed by remote processing devices that are linked through a communication network. The order in which the method 200 is described is not intended to be construed as a limitation, and any number of the described method blocks can be combined in any order to implement the method 200, or an alternative method. Furthermore, the method 200 can be implemented in any suitable hardware, software, firmware, or combination thereof.
[0025] At step 202 of the method 200, the one or more hardware processors 102 are configured by the programmed instructions to receive an incident report. The incident report comprises a plurality of words. For example, the incident report can be “On February 1, 2014, at approximately 11:37 a.m., a 340 ft.-high guyed telecommunication tower, suddenly collapsed during upgrading activities. Four employees were working on the tower removing its diagonals. In the process, no temporary supports were installed. As a result of the tower ’s collapse, two employees were killed, and two others were badly injured”. Here, the words “working”, “collapsed”, “activities”, “removing”, “collapse”, “killed” and “injured” are example events.
[0026] At step 204 of the method 200, the one or more hardware processors 102 are configured by the programmed instructions to compute a plurality of features corresponding to each of the plurality of words associated with the incident report using a pretrained Bidirectional Encoder Representations from Transformers (BERT) model. The BERT makes use of Transformer, an attention mechanism that learns contextual relations between words (or sub-words) in a text. In its vanilla form, Transformer includes two separate mechanisms, an encoder that reads the text input and a decoder that produces a prediction for the task. Since BERT’s goal is to generate a language model, only the encoder mechanism is necessary. In an embodiment, the plurality of features corresponding to a word is of 768X1 dimension. For example, the plurality of features looks like [0.2, 0.17, -0.06, ……………,1.04].
[0027] At step 206 of the method 200, the one or more hardware processors 102 are configured by the programmed instructions to compute a plurality of reduced features based on the plurality of features using a Fully Connected Neural Network (FCNN). In an embodiment, the FCNN is implemented using linear activation function to map the 768X1 dimensional space to 100 dimensional space. Due to linear activation, the FCNN layer acts like a linear transformation of the high dimensional input vector to a lower dimensional input vector. For example, the reduced 100X1 dimension features looks like [0.3, 0.14, -0.08, …,1.61].
[0028] At step 208 of the method 200, the one or more hardware processors 102 are configured by the programmed instructions to obtain a named entity information for each of the plurality of words using a Natural Language Processing (NLP) tool. In an embodiment, spaCy model has been used to classify named entities in text into pre-defined categories such as the names of persons, organizations, locations, expressions of times, quantities, monetary values, percentages. For example, the named entity information for each of the plurality of words is given as follows: The/O helicopter/O owned/ O by/ O Pawan/ B-ORG Hans/ I-ORG Limited/ I-ORG was and the like. Here, B-ORG indicates “beginning of an organization”. I-ORG indicates “inside an organizational entity but not the beginning one”. Hence Hans is assigned with the tag B-ORG. O indicates “other”.
[0029] At step 210 of the method 200, the one or more hardware processors 102 are configured by the programmed instructions to obtain a Parts Of Speech (POS) information for each of the plurality of words using the NLP tool. In an embodiment, the spaCy model has been used to obtain the POS information corresponding to each of the plurality of words. For example, the POS information for each of the plurality of words is given as follows: The/DT helicopter/NN owned/VBD by/IN Pawan/NNP Hans/NNP Limited/NNP was/VBD to/TO fly/VB from/IN Itanagar/NNP to/TO Tezpur/NNP. Here DT means determiner, NN means noun, IN means preposition, NNP means proper noun, VBD means past tense verb and the like.
[0030] At step 212 of the method 200, the one or more hardware processors 102 are configured by the programmed instructions to obtain a plurality of feature vectors corresponding to each of the plurality of words by concatenating the plurality of reduced features, the named entity information and the POS information. In an embodiment, the concatenated feature vector includes 140 features. The 140 features includes 100 reduced features obtained from the FCNN, 20 features corresponds to named entity information and the remaining 20 features corresponds to POS information. The concatenated features would be an array of 140 real numbers. For example, a sample array may be like [0.1, -0.12, 1.2, … ,0.7, ~ 0.3, 0.14, -0.08, … ,1.61, ~ 0.8, 0.12, 1.3, …, -0.09].
[0031] At step 214 of the method 200, the one or more hardware processors 102 are configured by the programmed instructions to compute a plurality of contextual feature values corresponding to each of the plurality of words based on the plurality of feature vectors using a pretrained Bidirectional Long Short Term Memory (BiLSTM), wherein the BiLSTM is pretrained using transfer learning method.
[0032] In an embodiment, the BiLSTM is trained using transfer learning method. Initially the BiLSTM is trained based on a general purpose annotated dataset. Further, the BiLSTM (trained with general purpose annotated dataset) is finetuned based on a small annotated industrial incident dataset.
[0033] In an embodiment, the method of training BiLSTM based on the general purpose annotated dataset includes the following steps. Initially, the general-purpose annotated dataset comprising a first set of annotated words is received. Further, a first set of features corresponding to each word from the first set of annotated words associated with the general-purpose annotated dataset are computed using a BERT model. Further, a first set of reduced features is computed based on the first set of features using the Fully Connected Neural Network (FCNN). Finally, the BiLSTM is trained based on the first set of reduced features and the first set of annotated words.
[0034] In an embodiment, the method of finetuning the BiLSTM based on the small annotated industrial incident dataset includes the following steps. Initially, the small annotated industrial incident dataset comprising a second set of annotated words is received. Further, a second set of features corresponding to each word from of the second set of annotated words are computed using the BERT model. Further, a second set of reduced features are computed based on the second set of features using the Fully Connected Neural Network (FCNN). After obtaining the reduced features, the BiLSTM is training using the second set of reduced features and the second set of annotated words.
[0035] At step 216 of the method 200, the one or more hardware processors 102 are configured by the programmed instructions to annotate each of the plurality of words with a corresponding label based on the plurality of contextual feature values using a pretrained Conditional Random Fields (CRF) model. The CRF is a statistical modeling method often applied in pattern recognition and machine learning and used for structured prediction. The CRF predicts a label for a single sample while considering the context from “neighbouring” samples. For example, event labels predicted by the CRF in the example is provided after a slash. The/O helicopter/O owned/O by/O Pawan/O Hans/O Limited/O flew/B-EVENT from/O Itanagar/O base/O but developed/B-EVENT a/O snag/B-EVENT midway/O ./O. The events are marked using the standard BIO (Begin-Inside-Outside) scheme for sequence labelling. For example, B-EVENT indicates beginning of an event (e.g., “snag/B-EVENT”) and O indicates any other non-event word (e.g., “helicopter/O”).
[0036] At step 218 of the method 200, the one or more hardware processors 102 are configured by the programmed instructions to identify a plurality of events from the annotated plurality of words based on the corresponding label. For example, in the plurality of words “The helicopter owned by Pawan Hans Limited flew from Itanagar base but developed a snag midway.”, the events are identified as “flew”, “developed” and “snag” based on the predicted labels.
[0037] At step 220 of the method 200, the one or more hardware processors 102 are configured by the programmed instructions to validate the identified plurality of events based on the plurality of contextual feature values using a clustering based validation technique.
[0038] In an embodiment, the clustering based validation technique for validating the identified plurality of events is explained below. Initially, the plurality of contextual feature values are received from the pretrained BiLSTM. Simultaneously a label corresponding to each of the plurality of contextual feature values are received from the CRF model. Further, a plurality of feature clusters are generated by clustering each of the plurality of contextual feature values using a complete agglomerative clustering. Further, for each feature cluster, each label corresponding to each of the plurality of contextual feature values are compared with a plurality of standard event labels. A flag corresponding to each feature cluster is set to one if the label corresponding to each of the plurality of contextual feature values are similar. Finally, the validation flag is set to one if the flag corresponding to each of the plurality of feature clusters is set to one. The event identification is considered as valid only if the validation flag is set to one.
[0039] For example, while clustering the BiLSTM feature values (the plurality of contextual features), a set of 10 feature clusters are formed. Each cluster is checked for the label corresponding to the feature values in the cluster. It is observed that in clusters 1 to 8, each label is EVENT and is a correct label when compared to the standard event dataset. Hence the flag of 1 is set to each of the 1 to 8 clusters. Further, it is observed that in clusters 9 and 10, each label is non-EVENT and is a correct label when compared to the standard event dataset. Hence a flag of 1 is assigned to clusters 9 and 10. In this example, since the flags corresponding each of the clusters and as they are one, the validation flag is also set to one to validate the event identification process.
[0040] FIG. 3 is an example process flow architecture (300) for the processor implemented method for transfer learning based event identification and validation implemented by the system of FIG. 1, in accordance with some embodiments of the present disclosure. Now referring to FIG.3, the process flow architecture is a sequential labelling architecture including a plurality of BERT layers 302A to 302N, a plurality of NER layers 304A to 304N, a plurality of POS layers 306A to 306N, a plurality of FCNN layers 308A to 308N, a plurality of concatenation layers 310A to 310N, a plurality of BiLSTM layers 312A to 312N and a CRF layer 314. The input incident report comprising the plurality of words is given as input. Each word is processed by the sequence of layers including the BERT layer, the NER layer, the POS layer, the FCNN layer, the concatenation layer, the BiLSTM layer and the CRF layer. The CRF layer outputs a label corresponding to each of the plurality of words. Initially, the word (w1) is given as input to the BERT layer 302A, the NER layer 304A and the POS layer 306A. The BERT layer 302A computes the plurality of features and the plurality of features are reduced by the FCNN layer 308 to obtain the plurality of reduced features. The NER layer 304A provides the named entity information and the POS layer 306A provides the POS information. Further, the plurality of reduced features, the named entity information and the POS information are concatenated by the concatenation layer 310A. The concatenated feature vector is given as input to the BiLSTM layer 312A which computes a plurality of contextual feature values. The plurality of contextual feature values are provided as input to the CRF layer 314 which generates the label for the word w1.
[0041] FIG. 4 is an example process flow architecture (400) for the clustering based validation technique associated with the processor implemented method for transfer learning based event identification and validation implemented by the system of FIG. 1, in accordance with some embodiments of the present disclosure. Now referring to FIG. 4, the process flow architecture includes an agglomerative clustering module 402 which receives the plurality of contextual features from the BiLSTM layers of FIG. 3 and clusters the plurality of contextual features into the plurality of feature clusters. The plurality of feature clusters are given as input to the matching module 406 which compares the label corresponding to each of the plurality of contextual features from each feature cluster with the plurality of standard events stored in a standard event label repository 404. A flag is set to one for each feature cluster if each label associated with each of the plurality of contextual features in the corresponding feature cluster is similar. The flag corresponding to each feature cluster is given as input to the validation module 408 which validates the system only if the flag value corresponding to each of the plurality of feature clusters are set to one.
[0042] In an embodiment, the system 100 is experimented as follows: The neural network architectures were developed in python using the keras package and the output has been evaluated using the seqeval package. For representing the text tokens as input in the proposed neural network approaches, the system was experimented with contextual embeddings (BERT and RoBERTa) available as part of the spacy_transformers package. Further, a 5-fold cross-validation was performed on the smaller incident training dataset for tuning the hyperparameters of the neural network.
[0043] The written description describes the subject matter herein to enable any person skilled in the art to make and use the embodiments. The scope of the subject matter embodiments is defined by the claims and may include other modifications that occur to those skilled in the art. Such other modifications are intended to be within the scope of the claims if they have similar elements that do not differ from the literal language of the claims or if they include equivalent elements with insubstantial differences from the literal language of the claims.
[0044] The embodiments of present disclosure herein address the unresolved problem of validating an identified event from an incident report. Here, the system is trained using transfer learning and the system utilized a linearly activated FCNN to reduce the dimensionality of the BERT features. Further, the system includes a clustering based validation technique which validates the performance of the system.
[0045] It is to be understood that the scope of the protection is extended to such a program and in addition to a computer-readable means having a message therein such computer-readable storage means contain program-code means for implementation of one or more steps of the method when the program runs on a server or mobile device or any suitable programmable device. The hardware device can be any kind of device which can be programmed including e.g. any kind of computer like a server or a personal computer, or the like, or any combination thereof. The device may also include means which could be e.g. hardware means like e.g. an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or a combination of hardware and software means, e.g. an ASIC and an FPGA, or at least one microprocessor and at least one memory with software modules located therein. Thus, the means can include both hardware means and software means. The method embodiments described herein could be implemented in hardware and software. The device may also include software means. Alternatively, the embodiments may be implemented on different hardware devices, e.g. using a plurality of CPUs, GPUs and edge computing devices.
[0046] The embodiments herein can comprise hardware and software elements. The embodiments that are implemented in software include but are not limited to, firmware, resident software, microcode, etc. The functions performed by various modules described herein may be implemented in other modules or combinations of other modules. For the purposes of this description, a computer-usable or computer readable medium can be any apparatus that can comprise, store, communicate, propagate, or transport the program for use by or in connection with the instruction execution system, apparatus, or device. The illustrated steps are set out to explain the exemplary embodiments shown, and it should be anticipated that ongoing technological development will change the manner in which particular functions are performed. These examples are presented herein for purposes of illustration, and not limitation. Further, the boundaries of the functional building blocks have been arbitrarily defined herein for the convenience of the description. Alternative boundaries can be defined so long as the specified functions and relationships thereof are appropriately performed. Alternatives (including equivalents, extensions, variations, deviations, etc., of those described herein) will be apparent to persons skilled in the relevant art(s) based on the teachings contained herein. Such alternatives fall within the scope and spirit of the disclosed embodiments. Also, the words “comprising,” “having,” “containing,” and “including,” and other similar forms are intended to be equivalent in meaning and be open ended in that an item or items following any one of these words is not meant to be an exhaustive listing of such item or items, or meant to be limited to only the listed item or items. It must also be noted that as used herein and in the appended claims, the singular forms “a,” “an,” and “the” include plural references unless the context clearly dictates otherwise. Furthermore, one or more computer-readable storage media may be utilized in implementing embodiments consistent with the present disclosure. A computer-readable storage medium refers to any type of physical memory on which information or data readable by a processor may be stored. Thus, a computer-readable storage medium may store instructions for execution by one or more processors, including instructions for causing the processor(s) to perform steps or stages consistent with the embodiments described herein. The term “computer-readable medium” should be understood to include tangible items and exclude carrier waves and transient signals, i.e. non-transitory. Examples include random access memory (RAM), read-only memory (ROM), volatile memory, nonvolatile memory, hard drives, CD ROMs, DVDs, flash drives, disks, and any other known physical storage media.
[0047] It is intended that the disclosure and examples be considered as exemplary only, with a true scope of disclosed embodiments being indicated by the following claims.