NZ762583B2 - Systems and methods for cross-media event detection and coreferencing - Google Patents

Systems and methods for cross-media event detection and coreferencing

Info

Publication number
NZ762583B2
NZ762583B2 NZ762583A NZ76258318A NZ762583B2 NZ 762583 B2 NZ762583 B2 NZ 762583B2 NZ 762583 A NZ762583 A NZ 762583A NZ 76258318 A NZ76258318 A NZ 76258318A NZ 762583 B2 NZ762583 B2 NZ 762583B2
Authority
NZ
New Zealand
Prior art keywords
event
determining
social media
similarity
alert
Prior art date
Application number
NZ762583A
Other versions
NZ762583A (en
Inventor
Lucas Carstens
John Duprey
Jochen Leidner
Armineh Nourbakhsh
Tim Nugent
Zarko Panic
Fabio Petroni
Natraj Raman
Sameena Shah
Original Assignee
Financial & Risk Organisation Limited
Filing date
Publication date
Application filed by Financial & Risk Organisation Limited filed Critical Financial & Risk Organisation Limited
Priority claimed from PCT/US2018/050885 external-priority patent/WO2019055654A1/en
Priority claimed from US16/130,390 external-priority patent/US11061946B2/en
Publication of NZ762583A publication Critical patent/NZ762583A/en
Publication of NZ762583B2 publication Critical patent/NZ762583B2/en

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/10—Office automation; Time management
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/10—Office automation; Time management
    • G06Q10/101—Collaborative creation, e.g. joint development of products or services
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q30/00—Commerce
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
    • G06Q50/01—Social networking

Abstract

method of providing cross-media event linking may include: receiving, at a first input of an event coreferencing system, a stream of social media postings, and at a second input, a stream of news articles; generating, by the event coreferencing system, a first set of event representations representing events referenced by the social media postings, and a second set of event representations representing events referenced by the news articles; determining, by the event coreferencing system, that at least one of the social media postings references a same event referenced by at least one of the news articles, the determining including determining at least one similarity using data of at least one of the first set of event representations.

Claims (24)

WHAT IS CLAIMED IS:
1. A method of providing cross-media event linking, the method comprising: receiving, at a first input of an event coreferencing system, a stream of social media postings, and at a second input of the event coreferencing system, a stream of news 5 articles; generating, by the event coreferencing system, a first set of event representations representing events referenced by the social media postings, and a second set of event representations representing events referenced by the news articles; determining, by the event coreferencing system, that at least one of the social media 10 postings references a same event referenced by at least one of the news articles, the determining including: determining at least one similarity between the at least one of the social media postings and the at least one of the news articles, the determining including: determining a temporal similarity between at least one time extracted 15 from the at least one social media posting and at least one time extracted from the at least one news article; and determining a spatial similarity between at least one location extracted from the at least one social media posting and at least one location extracted from the at least one news article; 20 generating a feature vector composed from the determined at least one similarity, including the determined temporal and spatial similarities; and inputting the feature vector to a classifier that outputs a classification of whether the at least one of the social media postings references the same event referenced by the at least one of the news articles based on the input feature vector; and 25 transmitting, by an output of the event resolution system to a user system, an alert including at least one co-referenced event representation representing the event referenced by the at least one of the social media postings and the at least one of the news articles; controlling by the user system at least one component of the user system based on 30 the alert; wherein the controlling includes at least one of: operating a supply chain scheduling controller to schedule a supply chain delivery based on the alert; or operating a manufacturing system controller to power down a manufacturing 35 system component based on the alert.
2. The method of claim 1, wherein the co-referenced event representation includes at least one of: links to the at least one of the social media postings and at least one of the news articles, or the at least one of the social media postings and the at least one of the news articles. 5
3. The method of claim 1 or claim 2, wherein transmitting the alert includes at least one of: transmitting an email including the alert to the user system, transmitting a text message including the alert to the user system, or transmitting a feed including the alert to the user system.
4. The method of any one of claims 1 to 3, wherein transmitting the alert includes 10 transmitting the alert by an application programming interface (API) of the event coreferencing system.
5. The method of claim 4, wherein the API transmits the alert in response to a request by the user system.
6. The method of any one of claims 1 to 5, wherein the controlling further includes 15 operating a financial trading controller to execute a trade of a financial commodity based on the alert.
7. The method of any one of claims 1 to 6, wherein each event representation includes a plurality of attributes representing the corresponding event.
8. The method of claim 7, wherein each event representation includes a location of the 20 event, a time of the event, and an impact of the event.
9. The method of any one of claims 1 to 8, wherein the determining the at least one similarity includes determining at least one of: a person entity similarity between at least one person entity extracted from the at least one social media posting and at least one person entity extracted from the at least one 25 news article; an organization entity similarity between at least one organization entity extracted from the at least one social media posting and at least one organization entity extracted from the at least one news article; or a text similarity between a text of the at least one social media posting and a text of 30 the at least one news article.
10. The method of any one of claims 1 to 9, wherein the determining the spatial similarity between the at least one social media posting and the at least one news article includes: determining feature vectors for a social media cluster and the news article based on 5 locations extracted from the social media posting and news article, the social media cluster including the at least one social media posting; calculating similarities between each pair of such locations of the news article and the social media cluster using the feature vectors; and determining the spatial similarity as function of the determined similarities. 10
11. The method of any one of claims 1 to 10, wherein the determining the temporal similarity between the at least one social media posting and the at least one news article includes: determining feature vectors for temporal expressions extracted from the social media posting and the news article; and 15 determining the temporal similarity based on a minimum time difference between temporal expressions in the feature vectors for the social media posting and the news article.
12. The method of any one of claims 1 to 11, wherein the determining the at least one similarity includes determining a person entity similarity between the social media posting 20 and the news article by: determining sets of person entities extracted from social media posting and the news article; and determining a similarity between the sets of extracted persons for the social media posting and the news article. 25
13. The method of any one of claims 1 to 12, wherein the determining the at least one similarity includes determining an organization entity similarity between the social media posting and the news article, including by: determining sets of organization entities extracted from social media posting and the news article; and 30 determining a similarity between the sets of extracted organization entities for the social media posting and the news article.
14. The method of any one of claims 1 to 13, wherein the determining the at least one similarity includes determining a text similarity between the social media posting and the news article, including generating vectors for tokenized text of the social media posting and the news article based on word embeddings, and determining a similarity between the 5 determined vectors.
15. The method of any one of claims 1 to 14, wherein the generating the second set of event representations representing events referenced by the news articles includes: filtering out non-event related news articles; classifying a type of event referenced by the news articles using a feature vector 10 based on word embeddings for the news articles; determining candidate attributes of the event representation using natural language processing; determining a location attribute of the event representation, including classifying candidate locations using feature vectors based on the candidate attributes; 15 determining a time attribute of the event representation, including determining temporal expressions in the news article and applying a rule based model to select one of the temporal expressions as the time attribute; and determining an impact attribute of the event representation, including classifying pairs of numeric references of the candidate attributes and adjacent word sequences. 20
16. The method of any one of claims 1 to 15, wherein the generating the second set of event representations representing events referenced by the news articles includes determining an impact attribute by: determining sentences of the news article with tokens with a cardinal number part- of-speech tag representing numeric values; 25 generating word sequences in the vicinity of the cardinal token in the text by constructing n-grams from each side of the cardinal token within the sentence; generating a feature vector for each sequence based on one or more of: word embeddings for the sequence, a length of the sequence, a pre or post offset of the cardinal number token relative to the sequence, a part-of-speech for words in the sequence, entity 30 types of the word sequence, dependency tree relations of the word sequence; and classifying each generated pairs of numeric value and word sequence feature vector as either indicating a human impact or not.
17. The method of any one of claims 1 to 16, wherein the generating the second set of event representations representing events referenced by the social media postings includes: filtering out noise postings representing spam and chit chat; detecting and clustering postings referencing an event; 5 classifying the type of event referenced by the cluster of postings; removing clusters of postings related to events older than a predetermined current time period; determining a summary for the cluster of postings; determining a location attribute for the cluster of postings; 10 determining a time attribute for the cluster of postings; and determining an impact attribute for the cluster of postings.
18. A system for providing cross-media event linking, the system comprising: at least one non-transitory machine readable storage medium storing program instructions; and 15 at least one processor configured to execute the program instructions to perform a method of providing cross-media event linking, the method including: receiving, at a first input of an event coreferencing system, a stream of social media postings, and at a second input of the event coreferencing system, a stream of news articles; 20 generating, by the event coreferencing system, a first set of event representations representing events referenced by the social media postings, and a second set of event representations representing events referenced by the news articles; determining, by the event coreferencing system, that at least one of the social media postings references a same event referenced by at least one of the news 25 articles, the determining including: determining at least one similarity between the at least one of the social media postings and the at least one of the news articles, the determining including: determining a temporal similarity between at least one time extracted from the at least one social media posting and at least one time extracted from 30 the at least one news article; and determining a spatial similarity between at least one location extracted from the at least one social media posting and at least one location extracted from the at least one news article; generating a feature vector composed from the determined at least one similarity, including the determined temporal and spatial similarities; and inputting the feature vector to a classifier that outputs a classification of whether the at least one of the social media postings references the same event 5 referenced by the at least one of the news articles based on the input feature vector; and transmitting, by an output of the event resolution system to a user system, an alert including at least one co-referenced event representation representing the event referenced by the at least one of the social media postings and the at least one of the news articles; and 10 controlling by the user system at least one component of the user system based on the alert; wherein the controlling includes at least one of: operating a supply chain scheduling controller to schedule a supply chain delivery based on the alert; or 15 operating a manufacturing system controller to power down a manufacturing system component based on the alert.
19. The system of claim 18, wherein transmitting the alert includes at least one of: transmitting an email including the alert to the user system, transmitting a text message including the alert to the user system, or transmitting a feed including the alert to the user 20 system.
20. The system of claim 18 or claim 19, wherein transmitting the alert includes transmitting the alert by an application programming interface (API) to the user system, wherein the API transmits the alert in response to a request by the user system.
21. The system of any one of claims 18 to 20, the method further comprising controlling 25 by the user system at least one component of the user system based on the alert.
22. The system of any one of claims 18 to 21, wherein each event representation includes a plurality of attributes representing the corresponding event, including a location of the event, a time of the event, and an impact of the event.
23. The system of any one of claims 18 to 22, wherein the determining the at least one 30 similarity includes determining at least one of: an entity similarity between the at least one social media posting and the at least one news article, and a text similarity between the at least one social media posting and the at least one news article.
24. At least one non-transitory machine readable storage medium having program instructions, which when executed by at least one processor perform a method of providing cross-media event linking, the method including: receiving, at a first input of an event coreferencing system, a stream of social media 5 postings, and at a second input of the event coreferencing system, a stream of news articles; generating, by the event coreferencing system, a first set of event representations representing events referenced by the social media postings, and a second set of event representations representing events referenced by the news articles; 10 determining, by the event coreferencing system, that at least one of the social media postings references a same event referenced by at least one of the news articles, the determining including: determining at least one similarity between the at least one of the social media postings and the at least one of the news articles, the determining including: 15 determining a temporal similarity between at least one time extracted from the at least one social media posting and at least one time extracted from the at least one news article; and determining a spatial similarity between at least one location extracted from the at least one social media posting and at least one location extracted from the at 20 least one news article; generating a feature vector composed from the determined at least one similarity, including the determined temporal and spatial similarities; and inputting the feature vector to a classifier that outputs a classification of whether the at least one of the social media postings references the same event referenced 25 by the at least one of the news articles based on the input feature vector; and transmitting, by an output of the event resolution system to the user system, an alert including at least one co-referenced event representation representing the event referenced by the at least one of the social media postings and the at least one of the news articles; and 30 controlling by the user system at least one component of the user system based on the alert; wherein the controlling includes at least one of: operating a supply chain scheduling controller to schedule a supply chain delivery based on the alert; or operating a manufacturing system controller to power down a manufacturing system component based on the alert.
NZ762583A 2018-09-13 Systems and methods for cross-media event detection and coreferencing NZ762583B2 (en)

Applications Claiming Priority (4)

Application Number Priority Date Filing Date Title
US201762559079P 2017-09-15 2017-09-15
US201762579218P 2017-10-31 2017-10-31
PCT/US2018/050885 WO2019055654A1 (en) 2017-09-15 2018-09-13 Systems and methods for cross-media event detection and coreferencing
US16/130,390 US11061946B2 (en) 2015-05-08 2018-09-13 Systems and methods for cross-media event detection and coreferencing

Publications (2)

Publication Number Publication Date
NZ762583A NZ762583A (en) 2024-01-26
NZ762583B2 true NZ762583B2 (en) 2024-04-30

Family

ID=

Similar Documents

Publication Publication Date Title
Deshpande et al. Depression detection using emotion artificial intelligence
AU2017355420B2 (en) Systems and methods for event detection and clustering
US20190012629A1 (en) Team performance supervisor
Guellil et al. Social big data mining: A survey focused on opinion mining and sentiments analysis
Artzi et al. Predicting responses to microblog posts
Zhou et al. An unsupervised framework of exploring events on twitter: Filtering, extraction and categorization
KR101972179B1 (en) Automatic task extraction and calendar entry
Wang et al. Context-aware intent identification in email conversations
Frías-Blanco et al. Fast adaptive stacking of ensembles
US10216838B1 (en) Generating and applying data extraction templates
CN111104518A (en) System and method for building an evolving ontology from user-generated content
Krstajic et al. Getting there first: Real-time detection of real-world incidents on Twitter
US20190332620A1 (en) Natural language processing and artificial intelligence based search system
WO2016066228A1 (en) Focused sentiment classification
Agarwal et al. Frame semantic tree kernels for social network extraction from text
Sintaha et al. An empirical study and analysis of the machine learning algorithms used in detecting cyberbullying in social media
Jung et al. Enhanced Naive Bayes classifier for real-time sentiment analysis with SparkR
Qadir et al. Activity modeling in email
Wang et al. When do words matter? Understanding the impact of lexical choice on audience perception using individual treatment effect estimation
US20190340670A1 (en) Clustering system, method, and program, and recommendation system
CN108090193B (en) Abnormal text recognition method and device
Al Marouf et al. Looking behind the mask: A framework for detecting character assassination via troll comments on social media using psycholinguistic tools
US11301777B1 (en) Determining stages of intent using text processing
NZ762583B2 (en) Systems and methods for cross-media event detection and coreferencing
NZ762583A (en) Systems and methods for cross-media event detection and coreferencing