Patent Application
20230305549
This U.S. patent application claims priority under 35 U.S.C. § 119 to: India Application No. 202221015893, filed on Mar. 22, 2022. The entire contents of the aforementioned application are incorporated herein by reference.
The disclosure herein generally relates to the field of incident analysis, and, more particularly, to systems and methods for similarity analysis in incident reports using event timeline representations.
Industrial incidents, even though highly undesirable, are an unavoidable reality. These could be industrial safety incidents or incidents related to cybersecurity. It is observed that cost of the industrial incidents runs into multiple billion dollars per annum. More importantly, there is irreparable human cost due to fatalities and major injuries such as permanent disabilities. In cybersecurity incidents, there may be loss of sensitive data as well as reputation of an organization which impacts business of organizations. In most cases, incident reports summarizing the incidents as well as their investigation are maintained in incident document repositories. Enterprises, regulatory bodies as well as standards committees spend extensive efforts to analyze incidents, identify root causes, and suggest preventive actions for recurrence and conduct trainings. Conventionally, multiple representations and frameworks such as the 24Model, Fishbone diagram, the event sequence diagram are used to represent, visualize and analyze incidents, but they are carried out with manual analysis.
While a few conventional approaches provide processing of an incident report which can be useful for populating knowledge representations, analyzing repositories of such incident reports and identifying useful insights, best practices, remedial measures, precautionary steps, on a large scale efficiently, remains challenging. Also, existing systems for analyzing the incident reports do not focus on identifying similar incidents. Further, due to unstructured text data used to describe the incident reports, it is very challenging to identify similar incident reports.
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 processor implemented method is provided. The method comprising receiving, via one or more hardware processors, (i) one or more incident reports from a repository of incident reports stored in a system database and (ii) an incoming query incident report pertaining to one or more industrial domains, wherein each of the one or more incident reports comprises a plurality of sentences, and wherein each of the plurality of sentences comprises a plurality of events indicative of a disruption in a service operation; generating, via the one or more hardware processors, a timeline representation of the plurality of events comprised in the plurality of sentences of (i) the one or more incident reports and (ii) the incoming query incident report using a temporal relation identification technique, wherein the temporal relation identification technique comprising: classifying, using a deep learning based neural network classifier, each of the plurality of sentences in the one or more incident reports into one of a set of predefined categories to obtain a plurality of time ordered sentences, wherein the set of predefined categories includes a background, an incident, and a consequence; and determining, a time order of the plurality of events in each of the plurality of time ordered sentences based on one or more temporal markers; determining, via the one or more hardware processors, a similarity measure between the incoming query incident report and the one or more incident reports from the repository of incident reports stored in the system database using a similarity identification approach that utilizes the generated timeline representation of the plurality of events, wherein the similarity identification approach comprising: obtaining, a first set of incidents reports from the one or more incident reports from the repository of incident reports stored in the system database, wherein the first set of incident reports having a textual content based similarity with the incoming query incident report based on an indexing method; identifying a set of similar events between each of the first set of incident reports and the incoming query incident report by applying at least one of (i) a dynamic programming based approach that is modified to not enforce a verbatim match amongst the plurality of events and (ii) one or more similarity parameters on the timeline representation of a plurality of events comprised in each of the first set of incident reports and a plurality of events comprised in the incoming query incident report; determining, using an approximate matching technique, a similarity score for each component of the set of similar events between each of the first set of incident reports and the incoming query incident report; and identifying, based on the similarity score, a longest sequence of similar components of the set of similar events between each of the first set of incident reports and the incoming query incident report; and dynamically updating, via the one or more hardware processors, the system database by storing the timeline representation and corresponding similarity measure of the plurality of events comprised in the plurality of sentences of the one or more incident reports and the incoming query incident report.
In another aspect, a system is provided. The system comprising a memory storing instructions; one or more communication interfaces; and one or more hardware processors coupled to the memory via the one or more communication interfaces, wherein the one or more hardware processors are configured by the instructions to receive, (i) one or more incident reports from a repository of incident reports stored in a system database and (ii) an incoming query incident report pertaining to one or more industrial domains, wherein each of the one or more incident reports comprises a plurality of sentences, and wherein each of the plurality of sentences comprises a plurality of events indicative of a disruption in a service operation; generate, a timeline representation of the plurality of events comprised in the plurality of sentences of (i) the one or more incident reports and (ii) the incoming query incident report using a temporal relation identification technique, wherein the temporal relation identification technique comprising: classifying, using a deep learning based neural network classifier, each of the plurality of sentences in the one or more incident reports into one of a set of predefined categories to obtain a plurality of time ordered sentences, wherein the set of predefined categories includes a background, an incident, and a consequence; and determining, a time order of the plurality of events in each of the plurality of time ordered sentences based on one or more temporal markers; determine, a similarity measure between the incoming query incident report and the one or more incident reports from the repository of incident reports stored in the system database using a similarity identification approach that utilizes the generated timeline representation of the plurality of events, wherein the similarity identification approach comprising: obtaining, a first set of incidents reports from the one or more incident reports from the repository of incident reports stored in the system database, wherein the first set of incident reports having a textual content based similarity with the incoming query incident report based on an indexing method; identifying a set of similar events between each of the first set of incident reports and the incoming query incident report by applying at least one of (i) a dynamic programming based approach that is modified to not enforce a verbatim match amongst the plurality of events and (ii) one or more similarity parameters on the timeline representation of a plurality of events comprised in each of the first set of incident reports and a plurality of events comprised in the incoming query incident report; determining, using an approximate matching technique, a similarity score for each component of the set of similar events between each of the first set of incident reports and the incoming query incident report; and identifying, based on the similarity score, a longest sequence of similar components of the set of similar events between each of the first set of incident reports and the incoming query incident report; and dynamically update, the system database by storing the timeline representation and corresponding similarity measure of the plurality of events comprised in the plurality of sentences of the one or more incident reports and the incoming query incident report.
In yet another aspect, a non-transitory computer readable medium is provided. The non-transitory computer readable medium comprising receiving, via one or more hardware processors, (i) one or more incident reports from a repository of incident reports stored in a system database and (ii) an incoming query incident report pertaining to one or more industrial domains, wherein each of the one or more incident reports comprises a plurality of sentences, and wherein each of the plurality of sentences comprises a plurality of events indicative of a disruption in a service operation; generating, via the one or more hardware processors, a timeline representation of the plurality of events comprised in the plurality of sentences of (i) the one or more incident reports and (ii) the incoming query incident report using a temporal relation identification technique, wherein the temporal relation identification technique comprising: classifying, using a deep learning based neural network classifier, each of the plurality of sentences in the one or more incident reports into one of a set of predefined categories to obtain a plurality of time ordered sentences, wherein the set of predefined categories includes a background, an incident, and a consequence; and determining, a time order of the plurality of events in each of the plurality of time ordered sentences based on one or more temporal markers; determining, via the one or more hardware processors, a similarity measure between the incoming query incident report and the one or more incident reports from the repository of incident reports stored in the system database using a similarity identification approach that utilizes the generated timeline representation of the plurality of events, wherein the similarity identification approach comprising: obtaining, a first set of incidents reports from the one or more incident reports from the repository of incident reports stored in the system database, wherein the first set of incident reports having a textual content based similarity with the incoming query incident report based on an indexing method; identifying a set of similar events between each of the first set of incident reports and the incoming query incident report by applying at least one of (i) a dynamic programming based approach that is modified to not enforce a verbatim match amongst the plurality of events and (ii) one or more similarity parameters on the timeline representation of a plurality of events comprised in each of the first set of incident reports and a plurality of events comprised in the incoming query incident report; determining, using an approximate matching technique, a similarity score for each component of the set of similar events between each of the first set of incident reports and the incoming query incident report; and identifying, based on the similarity score, a longest sequence of similar components of the set of similar events between each of the first set of incident reports and the incoming query incident report; and dynamically updating, via the one or more hardware processors, the system database by storing the timeline representation and corresponding similarity measure of the plurality of events comprised in the plurality of sentences of the one or more incident reports and the incoming query incident report.
In accordance with an embodiment of the present disclosure, the incoming query incident report is different from the one or more incident reports from the repository of incident reports stored in the system database.
In accordance with an embodiment of the present disclosure, the timeline representation is indicative of chronological ordering of the plurality of events.
In accordance with an embodiment of the present disclosure, the set of similar events is a subset of the timeline representation of a plurality of events comprised in each of the first set of incident reports and a plurality of events comprised in the incoming query incident report.
In accordance with an embodiment of the present disclosure, the one or more similarity parameters include one or more linguistic constraints, one or more word embedding constraints, and one or more sentence representations.
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.
The accompanying drawings, which are incorporated in and constitute a part of this disclosure, illustrate exemplary embodiments and, together with the description, serve to explain the disclosed principles.
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 scope of the disclosed embodiments.
Industrial incidents, even though highly undesirable, are an unavoidable reality. These could be industrial safety incidents or incidents related to cybersecurity. It is observed that cost of the industrial incidents runs into multiple billion dollars per annum. More importantly, there is irreparable human cost due to fatalities and major injuries such as permanent disabilities. In cybersecurity incidents, there may be loss of sensitive data as well as reputation of an organization which impacts business of organizations. In most cases, incident reports summarizing the incidents as well as their investigation are maintained in incident document repositories. Enterprises, regulatory bodies as well as standards committees spend extensive efforts to analyze incidents, identify root causes, suggest preventive actions for recurrence and conduct trainings. Conventionally, multiple representations and frameworks such as the 24Model, Fishbone diagram, the event sequence diagram are used to represent, visualize and analyze incidents, but they are carried out with manual analysis.
Although few conventional approaches provide processing of an incident report which can be useful for populating knowledge representations, but analyzing repositories of such incident reports and identifying useful insights, best practices, remedial measures, precautionary steps, on a large scale efficiently, remains challenging. Also, existing systems for analyzing the incident reports do not focus on identifying similar incidents. Further, unstructured text data is used to describe the incident reports. These challenges make it difficult for conventional approaches to be employed for identifying similar incident reports.
The present disclosure addresses unresolved problems of identifying similar events by automated analysis of a repository of incident reports using event timeline representations. Embodiments of the present disclosure provide systems and methods for similarity analysis in incident reports using event timeline representations. Timeline representation is an important knowledge representation which capture chronological ordering of the events. A timeline representation becomes useful in the process of root cause analysis as causes would temporally precede the effect (the incident in this case). To construct event timeline representations, chronological ordering of the events is required. The present disclosure provides a temporal relation identification technique to obtain a timeline representation of the events. Further, a similarity identification approach is used that makes use of neural embeddings to identify similar timeline representations and in turn, similar incident reports. More specifically, the present discourse provides:
Referring now to the drawings, and more particularly to
The I/O interface device(s) 106 can include a variety of software and hardware interfaces, for example, a web interface, a graphical user interface, and the like and can facilitate multiple communications within a wide variety of networks N/W 5 and protocol types, including wired networks, for example, LAN, cable, etc., and wireless networks, such as WLAN, cellular, or satellite. In an embodiment, the I/O interface device(s) can include one or more ports for connecting a number of devices to one another or to another server.
The memory 102 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, a system database 108 is comprised in the memory 102, wherein the system database 108 comprises one or more incident reports from a repository of incident reports, and timeline representation of events in the one or more incident reports. The database 108 further stores information on the scene in the environment. In an embodiment, the system database is dynamically updated
The system database 108 further comprises one or more networks such as one or more neural network(s) including deep learning based neural network classifier which when invoked and executed perform corresponding steps/actions as per the requirement by the system 100 to perform the methodologies described herein. The memory 102 further comprises (or may further comprise) information pertaining to input(s)/output(s) of each step performed by the systems and methods of the present disclosure. In other words, input(s) fed at each step and output(s) generated at each step are comprised in the memory 102 and can be utilized in further processing and analysis.
Referring to
In an embodiment, at step 204 of the present disclosure, the one or more hardware processors 104 are configured to generate, a timeline representation of the plurality of events comprised in the plurality of sentences of (i) the one or more incident reports and (ii) the incoming query incident report using a temporal relation identification technique. In an embodiment, the temporal relation identification technique comprising classifying, each of the plurality of sentences in the one or more incident reports into one of a set of predefined categories using a deep learning based neural network classifier to obtain a plurality of time ordered sentences. In an embodiment, the set of predefined categories includes a background, an incident, and a consequence; and determining, a time order of the plurality of events in each of the plurality of time ordered sentences based on one or more temporal markers. In an embodiment, the one or more temporal markers may include but not limited to temporal cues occurring as lexical markers such as after, before, when, and/or the like. In an embodiment, the timeline representation is indicative of chronological ordering of the plurality of events.
The steps 202 and 204 are better understood by way of the following description provided as exemplary explanation.
An incident occurs under some circumstances and would comprise of a series of undesirable events, finally ending with some aftermath either serious (i.e., involving injury or death or instrument damage) or mild. A sentence in the incident report gives information on one of the three aspects of the incident: pre-incident background circumstances (referred to as BACKGROUND), description of the incident (referred to as INCIDENT) and a post-incident aftermath (referred to as CONSEQUENCES). Table 1 below provides two examples of sample incident reports.
Sample incident #1 report from Table 1 is shown with each sentence marked with the corresponding aspect in below Table 2. As can be seen in Table 1 and Table 2, the first sentence which mentions the collapse of the communication tower is describing the incident. The next two sentences, however, give a description of events that were in progress just before the collapse such as the ongoing removal of the diagonals. The final sentence describes the aftermath involving death and injury to the employees.
Thus, while recording the incident, the incident report could be prepared to formulate a sentence to encode information about one or more of the three aspects—BACKGROUND, INCIDENT and CONSEQUENCES. However, to describe the incident, these aspects (i.e., the corresponding sentences) may not appear chronologically in textual content of the incident report.
The method of the present disclosure provides a temporal relation identification technique for temporal ordering of events (alternately referred to as ‘event temporal ordering’) that makes use of this observation. The task of event temporal ordering is divided into two steps: inter-sentence ordering and intra-sentence ordering. In the inter-sentence ordering step, each sentence in the incident reports is classified in one of the three aspects—BACKGROUND, INCIDENT and CONSEQUENCES. A deep learning based neural network classifier is used to perform the classification of each sentence to its corresponding aspect. As output of this step, a list of sentences ordered temporally is obtained by placing the BACKGROUND aspect sentences first, followed by the INCIDENT aspect sentences and then the CONSEQUENCES aspect sentences.
In an embodiment, once the sentences are classified into these ordered sentence pre-defined categories (i.e., BACKGROUND, INCIDENT, CONSEQUENCE categories), the plurality of events inside each sentence need to be ordered which is carried out as part of the intra-sentence ordering step. In order to achieve the classification of the events, the method of present disclosure harnesses few temporal cues occurring as lexical markers (say, after, before, when) to place intra-sentence events in correct order. The process of intra-sentence ordering involves checking a presence of temporal cues or lexical markers on lowest common ancestor path between the plurality of events and checking their occurrence with respect to a first event in a pair. In an embodiment, events are generally expressed in natural language as verbs. However, there are some events which are expressed as nouns such as attack, acquisition, and/or the like. The events expressed as nouns are known as nominal events. In the present disclosure, the nominal events are also handled by establishing a simultaneous relation with their dependency parent verbs which are verb-based events.
The entire process of intra sentence event ordering can be further better understood by way of following pseudo code provided as example:
Input: A set of events E for a sentence S, a set of Part of Speech (POS) tags P for sentence S and a set of dependency relations D for sentence S Result: Temporally ordered list of the input events OE_LIST
‘A power line crew was replacing a broken utility pole. One of the employees on the crew was standing on the side of the digger derrick holding onto the side of the truck with both hands. A second employee was reaching into the cab of the truck with his right knee touching the running board. The ground was wet from a rain and ice storm. As he was positioning the boom of the derrick the operator brought the boom into contact with a 13.2-kilovolt overhead power line. The employee at the side of the truck was electrocuted. The employee reaching into the truck received an electric shock and burns on his knee and foot. He was hospitalized for his injuries.’
In an embodiment, at step 206 of the present disclosure, the one or more hardware processors 104 are configured to determine a similarity measure between the incoming query incident report and the one or more incident reports from the repository of incident reports stored in the system database using a similarity identification approach that utilizes the generated timeline representation of the plurality of events. In an embodiment, the similarity identification approach could be alternatively referred as event timeline representations based similarity identification approach. Identifying incidents with similar causes or similar consequences or both can help in devising strategies to curb incidents and better post-incident remedial measures. However, to derive such minute observations, simple inverted index and query based search of similar incidents do not suffice. In addition to index based searching, finding similar event timelines help in establishing and observing fine-grained facets of incident similarity. However, task of finding similar timelines is not straightforward even for human experts owing to complex nature of the timeline representations. To obtain a quantitative measure of similarity between two event timeline representations, an objective definition of the similarity is required to be devised which should allow for an ordinal grading instead of a binary similar/not-similar selection. The classification of sentences into the set of predefined categories (i.e., Background, Incidents, Consequences) become defining dimensions if a comparison needs to be carried out between any two incidents. Similar backgrounds may indicate similar prevalent conditions and possibly causes. Similar incident descriptions may indicate similar series of failure events. More interestingly, similar incident descriptions but dissimilar backgrounds may indicate scenarios with different causes but similar incidents. Similarly, more such scenarios can be deciphered. More specifically, the method of the present disclosure performs similarity analysis in light of following points:
In an embodiment, the similarity identification approach, at first step, obtain, a first set of incidents reports from the one or more incident reports from the repository of incident reports stored in the system database. Here the first set of incidents reports refer to the incident reports having a textual content based similarity with the incoming query incident report based on an indexing method. In an embodiment, the first set of incidents are indicative of a set of k most similar incidents (alternatively referred as matching incidents). In other words, the method of the present disclosure finds similar incidents through a combination of (a) an inverted index based similarity, and (b) similarity in event timelines of the incidents. As the first step, an inverted index on a corpus of incidents and an incoming query incident is built. Further, the inverted index is queried to retrieve the set of k most similar incidents. A standard Information Retrieval (IR) pipeline of stop word removal and lemmatization is used for both corpus documents and the incoming query incident while creating and querying the inverted index respectively.
Further, at second step of the similarity identification approach, a set of similar events between each of the first set of incident reports and the incoming query incident report is identified by applying at least one of (i) a dynamic programming based approach that is modified to not enforce a verbatim match amongst the plurality of events and (ii) one or more similarity parameters on the timeline representation of a plurality of events comprised in each of the first set of incident reports and a plurality of events comprised in the incoming query incident report. In an embodiment, the verbatim match between two events refers to a match when two events are exactly same. For example, the events “The employee received an electric shock” and “The employee received an electric shock” are exactly matching. However, this is restrictive and the present disclosure allows for calling events as similar even if they are paraphrases of each other. For example, the events “The employee received an electric shock” and “The worker got a shock due to electricity” are considered as matching despite of having no exact match.
In an embodiment, the set of similar events is a subset of the timeline representation of a plurality of events comprised in each of the first set of incident reports and a plurality of events comprised in the incoming query incident report. In an embodiment, the one or more similarity parameters include one or more linguistic constraints, one or more word embedding constraints, and one or more sentence representations.
In other words, as the second step, an unsupervised timeline similarity matching is performed on the k most similar incidents obtained from the first step. The modified dynamic programming based approach is used to identify the longest sequence of similar events by unifying different sources of assessing similarity between events, namely, linguistic constraints, static word embeddings such as GloVe, and transformer based sentence representations such as SentBERT. In the context of the present disclosure, an event is defined as a four-tuple (e,A0,A1,EL) which includes an event phrase e, argument A0, argument A1 and complete event label EL. Arguments A0 and A1 represent semantic roles of an event, such that A0 represents an agent, doer or initiator of the event and A1 represents patient, the undergoer or affected entity of the event. The complete event label EL of an event is obtained by combining the event phrase e, argument A0 and argument A1 using syntactic dependencies. At first, a similarity between two events is determined and further the similarity between sequences within the similar events that are obtained based on event level mapping is determined. The similarity between two events is assessed based on the following similarity parameters:
Further, at third step of the similarity identification approach, a similarity score is determined for each component of the set of similar events between each of the first set of incident reports and the incoming query incident report using an approximate matching technique. At fourth step, a longest sequence of similar components of the set of similar events between each of the first set of incident reports and the incoming query incident report is identified based on the similarity score. This means that after measuring similarity between events, the longest sequence of similar events is identified. In the present disclosure, suppose two event timeline representation are shown as sequences of their corresponding events. Let E1=<e11,e12,e13, . . . , e1m> and E2=<e21,e22,e23, . . . , e2n> be two event sequences. Then, instead of comparing every event in E1 with every event in E2 through a brute-force approach, the longest non-contiguous similar sequence of events by computing this similarity recursively for a subset of events of two timelines. In an embodiment, the longest sequence could be non-contiguous and contiguous. For simplifying the explanation, the longest non-contiguous sequence is used in the present disclosure. In the present disclosure, an approximate matching process is devised based on the one or more similarity parameters such as the one or more linguistic constraints, the one or more word embedding constraints such as GloVe, and the one or more sentence representations such as SentBERT. Let C[m,n] represents length of the longest subsequence of E1 and E2. Here, C[m,n] can be computed based on C[i,j] such that i<m and j<n as shown in equation (1) below:
After computing values in Cmxn, the longest non-contiguous sequence of similar events can be obtained by parsing the matching points in Cmxn.
The entire similarity identification approach can be further better understood by way of following pseudo code provided as example:
Input: Event details: event1_details, event2_details
τ1=event phrase similarity threshold
τ2=event argument similarity threshold
φupper=upper bound threshold for SentBERT based similarity
φlower=lower bound threshold for SentBERT based similarity
antonym_list=pairs of words with antonym relations
Result: Similarity score
ev_1=event1_details[event]
dependencies_ev_1=event1_details[dependency_relations];
Similarly obtain A0_2, ev_2, A1_2, EL_2 and dependencies_ev_2;
events_sim=0.0;
events_sim_computed=False;
//Computation of events_sim is a complex process based on three phases
//Phase 1: Checking SentBERT based similarity
Sentbert_similarity=SentBERT_similarity(EL
Matching incident 2 text: ‘Two employees were working around a truck-mounted Hugh Williams 18.3-meter digger when the boom contacted a 7200-volt (phase-to-ground) overhead power line. One of the employees was electrocuted. The other worker received an electric shock for which he was hospitalized.’
Matching incident 3 text: ‘Two communications workers were realigning a telephone cable. As they were stringing a messenger cable for the telephone line between two poles the messenger cable contacted a 7.2-kilovolt overhead power line suspended from the top of the poles. One of the employees was electrocuted. The other employee received an electric shock and sustained only minor burns. (The second injured employee was not listed on an injury line on the original form.)’
Matching incident 4 text: ‘An employer lifted an employee in the bucket of a backhoe to untangle a Mylar balloon from an overhead power line. The bucket contacted the power line. The employee received an electric shock and fell from the bucket to the ground 3 meters below. He was hospitalized for his injuries.’
Matching incident 5 text: ‘A power line crew was replacing conductors on an overhead power line. Two power line workers on the crew were in the elevated bucket of an aerial lift. They were tightening a lighting arrester on one of the phases. One of the employees was holding the hot line clamp in one hand and tightening the arrester with a wrench in the other hand. The arrester twisted and the connector to the fusible disconnect hit the phase conductor that the employee was holding. The employee received an electric shock and was burned. He was hospitalized at a burn center for severe burns to both hands.’
As can be seen in
Referring to steps of
Datasets: In the present disclosure, performance of the method of the present disclosure is evaluated on two real life datasets. The datasets used in the method of present disclosure comprises aviation and construction incidents. For aviation dataset, summaries of aircraft incidents from the known in the art Skybrary repository (e.g., refer (https://www.skybrary.aero/index.php/Category:Accidents_and_Incidents) is crawled for multiple years, leading to a total of 1225 incidents. For construction, the dataset of 1863 Occupational Safety and Health Administration (OSHA) incident report summaries are used. In an embodiment, certain incidents as are selected as the incoming query incidents from these datasets and removed from rest of the corpus used for finding similar incidents. Considering the wide variety of incidents, the incoming query incidents are groped based on a common incident type. For aviation, 4 subsets of 10 queries each relating to waterbody crashes, object hits/strikes, equipment malfunctions and flight path excursions are collected. Similarly, for construction 4 subsets of 10 queries each relating to electrocution, worker falls, asphyxiation and vehicle related accidents are collected.
Evaluation: To validate this for incidents in the above datasets, the present disclosure first extracts the event timeline representations. In the present disclosure, the index search and a timeline similarity based matching between the query incidents and the corpus incidents is executed and a list of 5 best matching incidents with respect to a query incident is obtained. To obtain gold standard annotations for incident similarity, a Likert-scale based grading exercise of the 5 best incidents is set up for each of the 40 queries in the two datasets. As part of the grading exercise, each of the 5 result timelines for a query timeline is required to be annotated on a scale of 0 to 3, where 0 indicates no similarity and 3 indicates very high similarity.
In the present disclosure, a Normalized Discounted Cumulative Gain (NDCG) over the scores of the 5 results for each query is computed and an average over the scores for queries in each incident type is reported as shown in Table 3. Table 3 provides evaluation data for similarity analysis based on event timeline representations.
As a baseline, the method of the present disclosure is compared against a conventional approach which uses the exact match based Longest Common Subsequence (LCS) algorithm over the index search results. Further, the method of the present disclosure is not compared with any supervised baseline as no labelled training data on incident similarity is available. It can be observed from Table 1, that approximate nature of the method of present disclosure which enables embeddings based similarity, surpasses the performance of the state of the art exact matching based LCS algorithm in most of the incident types in both datasets.
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 herein 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 present disclosure if they have similar elements that do not differ from the literal language of the embodiments or if they include equivalent elements with insubstantial differences from the literal language of the embodiments described herein.
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 processing components 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.
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 components described herein may be implemented in other components or combinations of other components. 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 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., be 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.
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.
| Number | Date | Country | Kind |
|---|---|---|---|
| 202221015893 | Mar 2022 | IN | national |
