CN117094397A - Fine granularity event information extraction method, device and product based on shorthand - Google Patents
Fine granularity event information extraction method, device and product based on shorthand Download PDFInfo
- Publication number
- CN117094397A CN117094397A CN202311352557.7A CN202311352557A CN117094397A CN 117094397 A CN117094397 A CN 117094397A CN 202311352557 A CN202311352557 A CN 202311352557A CN 117094397 A CN117094397 A CN 117094397A
- Authority
- CN
- China
- Prior art keywords
- event
- sample
- argument
- short
- description
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000000605 extraction Methods 0.000 title claims abstract description 114
- 238000000034 method Methods 0.000 claims abstract description 40
- 238000012549 training Methods 0.000 claims description 51
- 238000002372 labelling Methods 0.000 claims description 33
- 238000001914 filtration Methods 0.000 claims description 27
- 239000002131 composite material Substances 0.000 claims description 24
- 238000004590 computer program Methods 0.000 claims description 17
- 150000001875 compounds Chemical class 0.000 claims description 12
- 238000010586 diagram Methods 0.000 description 11
- 239000000284 extract Substances 0.000 description 6
- 230000006870 function Effects 0.000 description 6
- 230000008569 process Effects 0.000 description 6
- 238000012545 processing Methods 0.000 description 5
- 238000004458 analytical method Methods 0.000 description 4
- 230000009286 beneficial effect Effects 0.000 description 4
- 238000001514 detection method Methods 0.000 description 4
- 230000009471 action Effects 0.000 description 3
- 230000000694 effects Effects 0.000 description 3
- 230000019771 cognition Effects 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 230000006698 induction Effects 0.000 description 2
- 230000010354 integration Effects 0.000 description 2
- 238000004519 manufacturing process Methods 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 230000004075 alteration Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 238000012805 post-processing Methods 0.000 description 1
- 230000000750 progressive effect Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N5/00—Computing arrangements using knowledge-based models
- G06N5/02—Knowledge representation; Symbolic representation
- G06N5/022—Knowledge engineering; Knowledge acquisition
- G06N5/025—Extracting rules from data
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/36—Creation of semantic tools, e.g. ontology or thesauri
- G06F16/367—Ontology
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/30—Semantic analysis
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management
Abstract
The application provides a method, a device and a product for extracting fine granularity event information based on shorthand, which relate to the technical field of knowledge engineering information extraction and comprise the following steps: extracting an event short argument and a trigger word corresponding to the event short argument from a text to be extracted by using a short argument extraction model, wherein the event short argument is an entity argument from which description information is removed in the event short argument; extracting the short meta description from the text to be extracted by using a short meta description extraction model; the short argument description represents description information of the event short argument; and matching and recursing the event shortleaf element, the trigger word corresponding to the event shortleaf element and the shortleaf element description to obtain a fine-granularity event information list.
Description
Technical Field
The application relates to the technical field of knowledge engineering information extraction, in particular to a method, a device and a product for extracting fine granularity event information based on shorthand.
Background
The event extraction is a key task in information extraction, namely extracting event information from natural language text and presenting the event information in a structured form for subsequent analysis application, and has wide application in the fields of automatic abstracting, automatic question answering, information retrieval and the like. In the event extraction process, event trigger word detection and argument detection are required, namely, the event type of the event and the event element corresponding to the event are judged.
However, the existing information extraction method is too simple, and cannot meet the fine-grained information extraction requirement for complex events, namely, for more complex event description texts, clear, accurate and specific event information is difficult to extract. Therefore, it is necessary to develop a method, a device and a product for extracting fine-granularity event information based on shorthand elements, so as to improve the effect of extracting event information, obtain more accurate and more clear event information, and support depth cognition and accurate induction of events.
Disclosure of Invention
In view of the foregoing, embodiments of the present application provide a method, apparatus, and product for extracting event information based on arguments, so as to overcome or at least partially solve the foregoing problems.
In a first aspect of the embodiment of the present application, a method for extracting fine granularity event information based on a shortargument is provided, including:
extracting an event short argument and a trigger word corresponding to the event short argument from a text to be extracted by using a short argument extraction model, wherein the event short argument represents an entity argument of which the descriptive information is removed from the event short argument;
extracting the short meta description from the text to be extracted by using a short meta description extraction model; the short argument description represents description information of the event short argument;
And matching and recursing the event shortleaf element, the trigger word corresponding to the event shortleaf element and the shortleaf element description to obtain a fine-granularity event information list.
In an alternative embodiment, the shorthand extraction model is trained by:
acquiring a sample text and a plurality of sample event arguments in the sample text;
splitting the sample event argument to obtain a sample event short argument, wherein the sample event short argument represents an entity argument directly related to the event;
labeling each sample event short element and sample trigger word in the sample text;
and training the pre-training model by using the marked sample text to obtain the shorthand meta-extraction model.
In an alternative embodiment, the splitting the sample event argument to obtain a sample event argument includes:
filtering sample event arguments which are irrelevant to the sample trigger words in the sample event arguments;
splitting and filtering remote entity description information in the sample event argument; the remote entity description information represents a position in a text and a existence distance of an entity, but is text information used for describing the entity in semantic information;
And under the condition that the sample event argument is a compound description type entity, splitting and filtering compound description information in the sample event argument to obtain the sample event argument, wherein the compound description type entity represents a text in which entity information and a plurality of description information exist in the form of a combination word.
In an alternative embodiment, the labeling each of the sample event shorthand and sample trigger words in the sample text includes:
labeling a sample trigger word, a related event and a short element label corresponding to each sample event short element in the sample text;
and labeling related events and trigger word labels corresponding to the sample trigger words for each sample trigger word in the sample text.
In an alternative embodiment, the short argument description extraction model is trained by:
acquiring a sample text;
defining a sample shorthand meta description and a sample shorthand meta entity in the sample text, wherein the sample shorthand meta description comprises a composite type description and a long-distance description;
marking the corresponding relation between each pair of the sample short meta description and the sample short meta entity in the sample text;
And training the pre-training model by using the marked sample text to obtain the short argument description extraction model.
In an alternative embodiment, said matching and recursing the event shorthand, the trigger word corresponding to the event shorthand, and the shorthand description includes:
filtering, from the short argument descriptions, short argument descriptions that are not related to the event short argument;
according to the event shortcuts, trigger words corresponding to the event shortcuts and the positions of the shortcuts in the text to be extracted, the filtered shortcuts are matched with the event shortcuts one by one, and a plurality of matching results are generated;
and recursion is carried out on the plurality of matching results according to the trigger words corresponding to the event shorthand meta-data to obtain the fine-granularity event information list.
In an optional implementation manner, the recursively obtaining the fine-grained event information list according to the trigger words corresponding to the event shorthand argument includes:
determining a plurality of related candidate matching results from the plurality of matching results according to trigger words corresponding to each event shorthand element;
Combining a plurality of candidate matching results to obtain one or more event information of the same event;
and generating the fine-granularity event information list according to the event information of each event.
A second aspect of an embodiment of the present application provides a device for extracting fine-granularity event information based on a shortargument, where the device includes:
the short argument extraction module is used for extracting event short arguments and trigger words corresponding to the event short arguments from the text to be extracted by utilizing a short argument extraction model, wherein the event short arguments represent entity arguments from which description information is removed;
the description extraction module is used for extracting the short meta description from the text to be extracted by using the short meta description extraction model; the short argument description represents description information of the event short argument;
and the event information generation module is used for matching and recursing the event shortleaf element, the trigger word corresponding to the event shortleaf element and the shortleaf element description to obtain a fine-granularity event information list.
In an alternative embodiment, the apparatus further includes a short argument extraction model training module, the short argument extraction model training module comprising:
A first sample acquisition sub-module for acquiring a sample text and a plurality of sample event arguments in the sample text;
the first definition sub-module is used for splitting the sample event argument to obtain a sample event short argument, and the sample event short argument represents an entity argument directly related to the event;
the first labeling sub-module is used for labeling each sample event shorthand meta-and sample trigger words in the sample text;
and the first training sub-module is used for training the pre-training model by using the marked sample text to obtain the shorthand extraction model.
In an alternative embodiment, the first defining sub-module includes:
the first filtering unit is used for filtering sample event arguments which are irrelevant to the sample trigger words in the sample event arguments;
the second filtering unit is used for splitting and filtering the remote entity description information in the sample event argument; the remote entity description information represents a position in a text and a existence distance of an entity, but is text information used for describing the entity in semantic information;
and the third filtering unit is used for splitting and filtering the composite description information in the sample event argument to obtain the sample event argument under the condition that the sample event argument is a composite description type entity, wherein the composite description type entity represents a text in which the entity information and a plurality of description information exist in the form of a combination word.
In an alternative embodiment, the first labeling sub-module includes:
the short-argument labeling unit is used for labeling the sample trigger word, the related event and the short-argument label corresponding to each sample event short argument in the sample text;
and the trigger word labeling unit is used for labeling related events and trigger word labels corresponding to the sample trigger words for each sample trigger word in the sample text.
In an alternative embodiment, the apparatus includes a short meta description extraction model training module comprising:
the second sample text acquisition sub-module is used for acquiring sample texts;
a second definition sub-module for defining a sample shorthand meta description and a sample shorthand meta entity in the sample text, the sample shorthand meta description including a composite type description and a remote description;
the second labeling sub-module is used for labeling the corresponding relation between each pair of the sample shorthand meta-description and the sample shorthand meta-entity in the sample text;
and the second training sub-module is used for training the pre-training model by using the marked sample text to obtain the short argument description extraction model.
In an alternative embodiment, the event information generation module includes:
a short argument description filtering sub-module for filtering short argument descriptions irrelevant to the event short argument from the short argument descriptions;
the matching sub-module is used for matching the filtered short-argument description with the event short-argument one by one according to the event short-argument, the trigger word corresponding to the event short-argument and the position of the short-argument description in the text to be extracted, and generating a plurality of matching results;
and the recursion sub-module is used for recursing the plurality of matching results according to the trigger words corresponding to the event shorthand argument to obtain the fine-granularity event information list.
In an alternative embodiment, the recursive sub-module comprises:
the determining unit is used for determining a plurality of related candidate matching results from the plurality of matching results according to the trigger words corresponding to each event short argument;
the combination unit is used for combining a plurality of candidate matching results to obtain one or more event information of the same event;
and the event information list generation unit is used for generating the fine-granularity event information list according to the event information of each event.
The third aspect of the embodiment of the application also provides an electronic device, which comprises a memory, a processor and a computer program stored on the memory, wherein the processor executes the computer program to realize the steps in the fine granularity event information extraction method based on the shortargument according to the first aspect of the embodiment of the application.
The fourth aspect of the embodiment of the present application further provides a computer readable storage medium, on which a computer program/instruction is stored, where the computer program/instruction implements the steps in the method for extracting fine granularity event information based on a argument according to the first aspect of the embodiment of the present application when the computer program/instruction is executed by a processor.
A fifth aspect of the embodiments of the present application also provides a computer program product, which when run on an electronic device causes a processor to perform the steps in the method for extracting fine-grained event information based on arguments according to the first aspect of the embodiments of the application.
The embodiment of the application provides a fine granularity event information extraction method based on shorthand, which comprises the following steps: extracting an event short argument and a trigger word corresponding to the event short argument from a text to be extracted by using a short argument extraction model, wherein the event short argument represents an entity argument of which the descriptive information is removed from the event short argument; extracting the short meta description from the text to be extracted by using a short meta description extraction model; the short argument description represents description information of the event short argument; and matching and recursing the event shortleaf element, the trigger word corresponding to the event shortleaf element and the shortleaf element description to obtain a fine-granularity event information list.
The concrete beneficial effects are that:
on one hand, the application extracts the event shorthand from the text to be extracted through the shorthand extraction model. Compared with the common event argument extracted according to the related technology, the extracted event shortargument removes the description information in the event argument, is the further splitting of the event argument, is beneficial to acquiring finer granularity argument information from the text of a complex event, and is beneficial to the subsequent integration of the event information.
On the other hand, the application extracts the short-argument description from the text to be extracted through the short-argument description extraction model, and then matches and recursively combines the short-argument description with the short-argument of the event, so that a complete event extraction result is generated, the extraction of remote and indirect description information from the text is facilitated, various description information and entity argument are connected, more complex argument descriptions are judged, more accurate event information with a clearer structure is obtained, and the deep cognition and accurate induction of the event are supported.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings that are needed in the description of the embodiments of the present application will be briefly described below, it being obvious that the drawings in the following description are only some embodiments of the present application, and that other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
FIG. 1 is a flowchart illustrating steps of a method for extracting fine granularity event information based on a argument according to an embodiment of the present application;
fig. 2 is a schematic diagram of a generation flow of a fine-grained event information list according to an embodiment of the application;
FIG. 3 is a schematic structural diagram of a fine granularity event information extraction device based on shortcuts according to an embodiment of the present application;
fig. 4 is a schematic diagram of an electronic device according to an embodiment of the present application.
Detailed Description
Exemplary embodiments of the present application will be described in more detail below with reference to the accompanying drawings in the embodiments of the present application. While exemplary embodiments of the present application are shown in the drawings, it should be understood that the present application may be embodied in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the application to those skilled in the art.
The event extraction is a key task in information extraction, namely extracting event information from natural language text and presenting the event information in a structured form for subsequent analysis application, and has wide application in the fields of automatic abstracting, automatic question answering, information retrieval and the like. In the event extraction process, event trigger word detection and argument detection are required, namely, the event type of the event and the event element corresponding to the event are judged. However, the existing information extraction method is too simple, and cannot meet the fine-grained information extraction requirement for complex events, namely, for more complex event description texts, clear and accurate event information is difficult to extract.
In complex events, not only the information in the form of (subject, trigger word, object) triples in the traditional event extraction and other entity information directly related to the trigger word, but also descriptive information of entities such as the subject, the object and the like are included. These indirect entity description information are fine-grained information of participation events, such as information of number description, model description, and country description of the entities. Meanwhile, the description of the entity may be complex, the descriptive nature may be far away from the entity, the same description may correspond to a plurality of entities, or one entity may have multiple descriptions and may be decomposed into a plurality of entities.
The traditional event extraction model method can not solve the problem of extracting fine granularity information of complex events, and mainly comprises the following three reasons:
(a) Indirection argument descriptions are difficult to judge for the event extraction model. The event extraction model firstly extracts the trigger words of the event and judges whether the argument and the trigger words are related or not, and for the argument description information indirectly forming the event, the argument description information is possibly similar to the argument description information not forming the event, so that the event extraction model cannot better judge the description information indirectly forming the event.
(b) The event extraction model cannot match the relation between different descriptions and entities, so that the extracted descriptive words can only match whether the extracted descriptive words are related to the trigger words or not, and the relation between the extracted descriptive words and the trigger words cannot be judged. If a single model is used to extract multiple meta-description information and entity information, the extraction model cannot link the multiple meta-description information and entity information, and only multiple individual description information and entity information can be extracted.
(c) The event extraction model can not judge the compound entity and description. For multiple described composite entities, the same description information may be multiplexed into multiple entities, and the event extraction model cannot determine such composite entities and descriptions.
Therefore, a single event extraction model can only find the relation between the argument and the trigger word, cannot judge the relation between the argument description and the trigger word, cannot match the argument with the argument description, cannot extract the composite argument description, cannot achieve an ideal event information extraction effect, and cannot extract accurate event information with clear structure for the text of a complex event.
In view of the above problems, embodiments of the present application provide a method, an apparatus, and a product for extracting fine-grained event information based on shorthand, so as to solve the problems of unsatisfactory event information extraction effects and the like. The method for extracting the fine granularity event information based on the shorthand metadata provided by the embodiment of the application is described in detail below through some embodiments and application scenes thereof with reference to the accompanying drawings.
Referring to fig. 1, fig. 1 is a step flowchart of a method for extracting fine-granularity event information based on a argument, as shown in fig. 1, where the method includes:
and step S101, extracting an event short argument and a trigger word corresponding to the event short argument from a text to be extracted by using a short argument extraction model, wherein the event short argument represents a physical argument from which description information is removed in the event short argument.
In the related art, the extraction of event information is often to directly extract event arguments (subject, object) and trigger words in text, and these triples are information. For the text describing the complex event, the normal long argument is often used as the extraction granularity to perform argument extraction, namely, the description information and the entity are extracted as a complete event argument, and the fine granularity description information cannot be obtained. Illustratively, illustrative sentences are extracted for one event: the A type and the B type automobiles collide with 2C type and D type trucks respectively, and the E type automobile is not impacted. According to the event information extraction method of the related art, it is possible to extract from the example sentence the event arguments "1 each of a type a and B type cars" as a subject, the trigger word "collision", and the event arguments "2 trucks of C type and D type" as objects. The 1-vehicle of the A-type and B-type vehicles extracted according to the method comprises part of descriptive information of A-type, B-type and 1-vehicle of the B-type vehicles, and for complex event information, the event theory is too long, so that the structure of the finally extracted event information is unclear and is not beneficial to the subsequent information analysis and processing.
The embodiment of the application provides an event short argument extracted from a text to be extracted by utilizing a pre-trained short argument extraction model. The text to be extracted represents text-form data containing descriptive information of the event. The event argument represents an entity argument from which the description information is removed. The event argument is an argument entity containing part of the descriptive information extracted in the related art, such as "1 car of a type and B type", and the event argument in this embodiment is that all the descriptive information is removed, and the entity information is reserved, which indicates an argument of an entity directly related to the event, such as "car". Optionally, the event short argument shows an event subject and an event object in the triplet information, and further includes time, place and other elements.
In this embodiment, the text to be extracted is input into the short argument extraction model, and the model performs analysis and calculation, and outputs an extraction result, where the extraction result includes an event short argument and a trigger word corresponding to the event short argument. Namely, for each extracted event short element, a corresponding trigger word can be extracted, so that according to the corresponding relation between the event short element and the trigger word, which event the event short element belongs to can be determined, and whether an entity is related to the trigger word or not is judged to filter out the entity irrelevant to the event. The embodiment of the application utilizes a short argument extraction model to extract and obtain the event short argument, and ignores descriptive information related to the event entity.
In an alternative embodiment, the shorthand extraction model is trained by:
step S201, a sample text and a plurality of sample event arguments in the sample text are acquired.
Step S202, splitting the sample event argument to obtain a sample event argument, wherein the sample event argument represents an entity argument directly related to the event. The sample text represents text data for describing the event, and common sample event arguments in the sample text represent arguments which refer to event subjects and objects and contain description information of entities. In the embodiment, the sample event argument is split, the description information in the sample event argument is removed, and the entity information in the sample event argument is reserved, so that a shorter sample event argument is obtained.
In an optional embodiment, the step S202 splits the sample event argument to obtain a sample event argument, including:
filtering sample event arguments which are irrelevant to the sample trigger words in the sample event arguments;
splitting and filtering remote entity description information in the sample event argument; the remote entity description information represents a position in a text and a existence distance of an entity, but is text information used for describing the entity in semantic information;
And under the condition that the sample event argument is a compound description type entity, splitting and filtering compound description information in the sample event argument to obtain the sample event argument, wherein the compound description type entity represents a text in which entity information and a plurality of description information exist in the form of a combination word.
In this embodiment, for a common sample event argument, the argument may be split according to actual requirements, and only a main short entity portion is reserved as a short argument of the event. Illustratively, for the sample event argument "1 car type a, B each", after splitting it, the short entity (sample event argument) that remains is "car". Specifically, the entity description that needs to be split and filtered from the sample event shorthand includes remote entity description information, i.e. there is no close connection to the entity part, and the entity is far away from the position in the text, but text information used for describing the entity is in semantic information. The present embodiment also needs to filter some description information in long entities, that is, for description type entities of composite descriptions that may occur, if there is a need for splitting, it also needs to filter. For example, for a compound descriptive entity "guest" where "guest" may be split as descriptive information of country type, the "guest" may be reserved as a sample event shorthand element according to actual extraction requirements. In addition, in the process, the embodiment also filters sample event arguments which are irrelevant to the sample trigger words in the sample event arguments. This is because entity information which is not related to the event itself except the event subject and the event object often exists in the text, and by judging the relation between the sample event argument and the sample trigger word, the argument which is not related to the trigger word can be determined as the entity information which is not related to the event itself, and filtered. For example, for the sample text "1 car of type a, B collides with 2 cars of type C, D, the car of type E is not impacted", the sample event argument "E car" is the sample event argument unrelated to the trigger word, which may be filtered out.
And step S203, labeling each sample event short argument and sample trigger word in the sample text.
After splitting the argument, each sample event short argument and trigger word in the sample text needs to be annotated according to the event. The label is different from the ordinary event label in that only short arguments and trigger words in the event need to be labeled, and no label is made for other descriptive argument information in the event.
In an optional implementation manner, the step S203 marks each of the sample event shorthand and the sample trigger word in the sample text, including:
and labeling the sample trigger word, the related event and the short element label corresponding to each sample event short element in the sample text. For example, for a sample event short-term "car" in a sample text, a sample trigger word "collision" corresponding to the short-term needs to be marked, so that an event related to the short-term can be determined according to the trigger word, and a short-term tag is used to mark that the text "car" belongs to the sample event short-term.
And labeling related events and trigger word labels corresponding to the sample trigger words for each sample trigger word in the sample text. For example, for a sample trigger word "collision" in a sample text, a related event corresponding to the word needs to be marked, generally, one trigger word corresponds to one event, and the trigger word label is used to mark that the text "collision" belongs to the sample trigger word.
And step S204, training the pre-training model by using the marked sample text to obtain the shorthand extraction model.
In this embodiment, the initial model is trained or the pre-trained model is fine-tuned according to the annotated sample text. Specifically, the sample text is used as training data, labels (sample event shortnotes and sample trigger words) in the sample text are used as labels, a pre-training model is trained, a loss value is calculated through a loss function, model parameter feedback is carried out according to the loss value, iteration training of the pre-training model is achieved through repeating the steps, training is finished after the preset training times are reached, and the shortnotes extraction model is obtained. Optionally, the pre-training model may be a pre-training language model T5, and the shorthand extraction may be performed by fine tuning using a relevant corpus based on the pre-training model T5, so as to obtain the shorthand extraction model.
According to the embodiment of the application, the sample event shortages and the trigger words in the sample text are marked, the marked sample text is used for training the model, so that a shortages extraction model is obtained, the event shortages and the trigger words can be extracted from the text to be extracted, and the entity irrelevant to the event can be filtered by judging whether the entity (event shortages) pre-trigger words are relevant or not. For example, for the text to be extracted, 1 of the type a and the type B automobiles collide with 2 of the type C and the type D trucks, the type E automobile is not impacted, and the result extracted from the text by the shorthand meta-extraction model can be expressed as: [ { event type: impact, subject: car, trigger word: impact, object: truck } ]. The extraction result comprises event shortages as event subjects and event objects, corresponding trigger words and event types, descriptive information related to event entities, such as model information of type A, type B and the like, is ignored, and only shortages directly related to events are extracted. In addition, in the extraction process, the shortleaf extraction model filters out entity information irrelevant to the trigger words, namely entity information relevant to the E-type automobile, so that more accurate event information can be obtained through subsequent integration.
Step S102, extracting short meta description from the text to be extracted by using a short meta description extraction model; the short argument description represents description information of the event short argument.
In this embodiment, the extracting model is described by using the shortnotes, and descriptive information about the shortnotes is extracted from the text to be extracted, for example, information about the number of event subjects, the country to which the event subjects belong, and the like. Exemplary, such as "type a", "type B", "1" each, and the like describe the event subject "car". The content extracted by the short-term description extraction model has no direct relation with the event, and is an extraction of description information aiming at short entities, so that the extraction result of the short-term description needs to be combined with the extraction result of the event short-term in the step S101 to generate a complete event extraction result.
In an alternative embodiment, the short argument description extraction model is trained by:
step S301, a sample text is acquired.
Step S302, defining a sample shorthand meta description and a sample shorthand meta entity in the sample text, wherein the sample shorthand meta description comprises a compound type description and a long-distance description.
In this embodiment, according to the actual requirement, the entity definition of the shorthand meta description is determined, that is, the sample shorthand meta description in the sample text is defined, and the sample shorthand meta entity corresponding to the sample shorthand meta description is determined. For example, for event subjects "type a, type B cars each 1" in the sample text, sample shorthand meta-descriptions "type a" and "type B" representing model description text and sample shorthand meta-descriptions "1" representing quantity description text may be determined, and it may be determined that the three sample shorthand meta-descriptions are for describing the sample shorthand entity of "car". It should be noted that the sample shorthand meta-description may be a composite type description or a remote description. Specifically, a combination of descriptions may be extracted for the composite short entity in the sample text, for example, two sample shorthand meta-descriptions numbered "001" and country category "mei" may be extracted from "001 mei guest". The remote description in the sample text indicates that there is no close connection with the entity part, is far from the entity in the text, but is text information for describing the entity in the semantic information.
Step S303, marking a correspondence between each pair of the sample shorthand meta descriptions and the sample shorthand meta entity in the sample text.
In this embodiment, after determining the sample short meta description and the sample short meta entity in the sample text, a pair of the sample short meta description and the sample short meta entity is labeled, specifically, for each sample short meta description in the sample text, the short meta entity of the description is labeled, and a description tag is added for the short meta entity to label that the text belongs to the sample short meta description; for each sample shortmeta entity in the sample text, marking one or more corresponding sample shortmeta descriptions, and adding an entity label for the sample shortmeta entity, wherein the label is used for marking that the text belongs to the sample shortmeta entity. For 1 automobile of the type A and the type B in the sample text, labeling the corresponding relation between the text of the type A and the text of the type B and the sample shortmeta entity of the automobile respectively, and adding a description label; for the text "car" therein, the correspondence relationship with the sample shorthand meta description "a type" and "B type" is noted, and an entity tag is added.
In this embodiment, not only the sample shorthand element entity and the corresponding sample shorthand element description related to the event need to be marked, but also the sample shorthand element entity and the corresponding sample shorthand element description unrelated to the event need to be marked, so that the subsequent model can train by using the marked text, and learn the matching information of the shorthand element entity and the shorthand element description. The difference between the labeling of the embodiment and the event labeling in the related technology is that in this step, only the shorthand meta-entity and the shorthand meta-description pair in the sample text need to be labeled, and no labeling is performed on other elements such as trigger words in the event.
And step S304, training the pre-training model by using the marked sample text to obtain the short meta description extraction model.
In this embodiment, the initial model is trained or the pre-trained model is fine-tuned according to the labeled corpus. Specifically, the sample text is used as training data, labels (paired sample shorthand meta description and sample shorthand meta entity) in the sample text are used as labels, a pre-training model is trained, a loss value is calculated through a loss function, model parameter feedback is carried out according to the loss value, iterative training of the pre-training model is achieved through repeating the steps, training is finished after the preset training times are reached, and a shorthand meta description extraction model is obtained. Optionally, the pre-training model may be a pre-training language model T5, and the extracting of the short meta description may be fine-tuned by using a related corpus based on the pre-training model T5, so as to obtain a short meta description extracting model, where the content extracted by the short meta description extracting model has no direct relation with the event, and the extracting result of the short meta description extracting model needs to be combined with the extracting result of the short meta extracting model to generate a complete event extracting result.
According to the embodiment of the application, each pair of sample short meta description and sample short meta entity in the sample text is marked, and the marked sample text is used for training the model, so that a short meta description extraction model is obtained, the short meta description can be extracted from the text to be extracted, the entity to which the short meta description belongs is determined, and the corresponding relation between the short meta description and the entity is obtained. For example, for the text to be extracted, that is, for 1 type a and type B car, 2 type C and type D cars collide, and for the text that type E car is not impacted, the extraction model is described by a short argument, and the result obtained by extraction from the extraction model can be expressed as:
[ { shorthand argument: automobile, shorthand argument model: A type, shorthand argument number: 1 },
{ number of arguments: car, number of arguments: type B, number of arguments: 1 },
{ number of arguments: truck, number of arguments: type C, number of arguments: 2 },
{ number of arguments: truck, number of arguments: D-type, number of arguments: 2 },
{ number of arguments: automobile, number of arguments: none } ].
The extraction result comprises event shortnotes serving as event subjects and event objects and corresponding shortnotes descriptions, and descriptive information related to entities irrelevant to the event, namely entity information related to 'E-type automobiles', and descriptive information can be extracted. It can be seen that this extraction is for short entities, so that a combination of descriptions can be extracted for one composite short entity. For multiple descriptions of one short composite entity, the description information of the short composite entity needs to be matched by a method of adding post-processing rules when the short composite entity has multiple groups of composite descriptions.
And step S103, matching and recursing the event shortcutter, the trigger word corresponding to the event shortcutter and the shortcutter description to obtain a fine-granularity event information list.
In the related art, when the description is extracted as an event element, it is difficult to match the relationship between the description and the entity, and description information and entity irrelevant to the event may be extracted, so as to obtain a plurality of pieces of individual description information and entity information. For example, for the text to be extracted that 1 type A and B cars collide with 2 types of C-type and D-type trucks, and the E-type car is not impacted, the extracted result can be expressed as [ { event type: collision, subject: car, trigger word: collision, object: truck, model: [ type A, type B, type C, type D, E ], number: [1, 2 ] } ].
Referring to fig. 2, fig. 2 shows a flow chart of generating a fine-granularity event information list, and as shown in fig. 2, this embodiment proposes that, in combination with the extraction result of the shorthand meta-extraction model in step S101 and the extraction result of the shorthand meta-description extraction model in step S102, matching and recursion are performed by a rule method, so as to generate a final fine-granularity event information list.
In an alternative embodiment, said matching and recursing the event shorthand, the trigger word corresponding to the event shorthand, and the shorthand description includes:
From the short argument descriptions, short argument descriptions that are not related to the event short argument are filtered. Specifically, since the description of the short meta-entity related to the event and the description of the entity unrelated to the event are included in the description of the short meta-entity extracted in step S102, the short meta-description information that is not in the event needs to be filtered out. According to the above example, the "E-type" related shorthand description is filtered out of the shorthand description, as its shorthand does not participate in the trigger word "collision" related event.
And according to the event shortcuts, trigger words corresponding to the event shortcuts and the positions of the shortcuts in the text to be extracted, performing one-to-one matching on the filtered shortcuts and the event shortcuts to generate a plurality of matching results.
In this embodiment, the matching needs to combine the short argument description with the event short argument according to the position of each short argument description in the text to be extracted and the relation between the short argument description extracted by the short argument description extraction model and the entity, so as to obtain a matching result. Illustratively, the shorthand meta-description located in the first sentence of the text is matched with the event shorthand meta-also located in the first sentence. It should be noted that multiple events may be included in the same sentence, which requires matching event arguments one by one with each event. Illustratively, according to the above example, the correspondence between the event shortcuts and the shortcuts description is determined, and the matching result is as follows:
[ { shorthand argument: automobile, shorthand argument model: A type, shorthand argument number: 1 },
{ number of arguments: car, number of arguments: type B, number of arguments: 1 },
{ number of arguments: truck, number of arguments: type C, number of arguments: 2 },
{ number of arguments: truck, number of arguments: type D, number of arguments: 2 } ].
And recursion is carried out on the plurality of matching results according to the trigger words corresponding to the event shorthand meta-data to obtain the fine-granularity event information list.
In an optional implementation manner, the recursively obtaining the fine-grained event information list according to the trigger words corresponding to the event shorthand argument includes:
determining a plurality of related candidate matching results from the plurality of matching results according to trigger words corresponding to each event shorthand element;
combining a plurality of candidate matching results to obtain one or more event information of the same event;
and generating the fine-granularity event information list according to the event information of each event.
In this embodiment, because one event argument may include a plurality of description information, and the same description information may be multiplexed into a plurality of entities, in the recursion process, a plurality of event arguments and a plurality of related argument descriptions need to be combined respectively to determine such a composite entity, description, so as to obtain complete event information. Specifically, a plurality of candidate matching results belonging to the same event can be determined according to the trigger word, and the matching results can be expressed as a combination of the event shortnotes and the shortnotes description. And then respectively combining a plurality of candidate matching results to obtain one or more event information. For example, for the text to be extracted, 1 of the vehicles a and B collide with 2 vehicles C and D, and the vehicle E is not impacted, according to the matching result obtained after matching, it can be known that the event subject is a composite subject, namely, 1 vehicle a and 1 vehicle B, and the event object is a composite object, namely, 2 vehicles C and 2 vehicles D, respectively, and the obtained multiple event information is as follows:
[ { event type: collision, number of subjects: 1, subject type: A, subject: automobile, trigger word: collision, number of objects: 2, object type: C, object: truck },
{ event type: collision, number of subjects: 1, subject type: A, subject: automobile, trigger word: collision, number of objects: 2, object type: D, object: truck },
{ event type: collision, number of subjects: 1, subject type: B, subject: automobile, trigger word: collision, number of objects: 2, object type: C, object: truck },
{ event type: collision, number of subjects: 1, subject type: B, subject: car, trigger word: collision, number of objects: 2, object type: D, object: truck } ].
And integrating the obtained event information to generate a complete fine-grained event information list.
The second aspect of the embodiment of the present application further provides a device for extracting fine-granularity event information based on a shorthand, referring to fig. 3, fig. 3 shows a schematic structural diagram of a device for extracting fine-granularity event information based on a shorthand, as shown in fig. 3, where the device includes:
the short argument extraction module is used for extracting event short arguments and trigger words corresponding to the event short arguments from the text to be extracted by utilizing a short argument extraction model, wherein the event short arguments represent entity arguments from which description information is removed;
The description extraction module is used for extracting the short meta description from the text to be extracted by using the short meta description extraction model; the short argument description represents description information of the event short argument;
and the event information generation module is used for matching and recursing the event shortleaf element, the trigger word corresponding to the event shortleaf element and the shortleaf element description to obtain a fine-granularity event information list.
In an alternative embodiment, the apparatus further includes a short argument extraction model training module, the short argument extraction model training module comprising:
a first sample acquisition sub-module for acquiring a sample text and a plurality of sample event arguments in the sample text;
the first definition sub-module is used for splitting the sample event argument to obtain a sample event short argument, and the sample event short argument represents an entity argument directly related to the event;
the first labeling sub-module is used for labeling each sample event shorthand meta-and sample trigger words in the sample text;
and the first training sub-module is used for training the pre-training model by using the marked sample text to obtain the shorthand extraction model.
In an alternative embodiment, the first defining sub-module includes:
the first filtering unit is used for filtering sample event arguments which are irrelevant to the sample trigger words in the sample event arguments;
the second filtering unit is used for splitting and filtering the remote entity description information in the sample event argument; the remote entity description information represents a position in a text and a existence distance of an entity, but is text information used for describing the entity in semantic information;
and the third filtering unit is used for splitting and filtering the composite description information in the sample event argument to obtain the sample event argument under the condition that the sample event argument is a composite description type entity, wherein the composite description type entity represents a text in which the entity information and a plurality of description information exist in the form of a combination word.
In an alternative embodiment, the first labeling sub-module includes:
the short-argument labeling unit is used for labeling the sample trigger word, the related event and the short-argument label corresponding to each sample event short argument in the sample text;
and the trigger word labeling unit is used for labeling related events and trigger word labels corresponding to the sample trigger words for each sample trigger word in the sample text.
In an alternative embodiment, the apparatus includes a short meta description extraction model training module comprising:
the second sample text acquisition sub-module is used for acquiring sample texts;
a second definition sub-module for defining a sample shorthand meta description and a sample shorthand meta entity in the sample text, the sample shorthand meta description including a composite type description and a remote description;
the second labeling sub-module is used for labeling the corresponding relation between each pair of the sample shorthand meta-description and the sample shorthand meta-entity in the sample text;
and the second training sub-module is used for training the pre-training model by using the marked sample text to obtain the short argument description extraction model.
In an alternative embodiment, the event information generation module includes:
a short argument description filtering sub-module for filtering short argument descriptions irrelevant to the event short argument from the short argument descriptions;
the matching sub-module is used for matching the filtered short-argument description with the event short-argument one by one according to the event short-argument, the trigger word corresponding to the event short-argument and the position of the short-argument description in the text to be extracted, and generating a plurality of matching results;
And the recursion sub-module is used for recursing the plurality of matching results according to the trigger words corresponding to the event shorthand argument to obtain the fine-granularity event information list.
In an alternative embodiment, the recursive sub-module comprises:
the determining unit is used for determining a plurality of related candidate matching results from the plurality of matching results according to the trigger words corresponding to each event short argument;
the combination unit is used for combining a plurality of candidate matching results to obtain one or more event information of the same event;
and the event information list generation unit is used for generating the fine-granularity event information list according to the event information of each event.
The embodiment of the application also provides an electronic device, and referring to fig. 4, fig. 4 is a schematic diagram of the electronic device according to the embodiment of the application. As shown in fig. 4, the electronic device 100 includes: the processor 120 is connected with the memory 110 through bus communication, and a computer program is stored in the memory 110 and can run on the processor 120, so that the steps in the fine granularity event information extraction method based on the shorthand element disclosed by the embodiment of the application are realized.
The embodiment of the application also provides a computer readable storage medium, on which a computer program/instruction is stored, which when executed by a processor, implements the steps in the method for extracting fine granularity event information based on shortfalls as disclosed in the embodiment of the application.
Embodiments of the present application also provide a computer program product which, when run on an electronic device, causes a processor to perform the steps of the method for extracting fine-grained event information based on shortfalls as disclosed in the embodiments of the present application.
In this specification, each embodiment is described in a progressive manner, and each embodiment is mainly described by differences from other embodiments, and identical and similar parts between the embodiments are all enough to be referred to each other.
Embodiments of the present application are described with reference to flowchart illustrations and/or block diagrams of methods, apparatus, electronic devices, and computer program products according to embodiments of the application. It will be understood that each flow and/or block of the flowchart illustrations and/or block diagrams, and combinations of flows and/or blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing terminal device to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing terminal device, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
While preferred embodiments of the present application have been described, additional variations and modifications in those embodiments may occur to those skilled in the art once they learn of the basic inventive concepts. It is therefore intended that the following claims be interpreted as including the preferred embodiment and all such alterations and modifications as fall within the scope of the embodiments of the application.
Finally, it is further noted that relational terms such as first and second, and the like are used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Moreover, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or terminal that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or terminal. Without further limitation, an element defined by the phrase "comprising one … …" does not exclude the presence of other like elements in a process, method, article or terminal device comprising the element.
The foregoing has described in detail the method, apparatus and article of manufacture for extracting fine-grained event information based on shorthand arguments, and specific examples have been used herein to illustrate the principles and embodiments of the application, the above examples being for the purpose of aiding in understanding the method and core concept thereof; meanwhile, as those skilled in the art will have variations in the specific embodiments and application scope in accordance with the ideas of the present application, the present description should not be construed as limiting the present application in view of the above.
Claims (10)
1. A method for extracting fine granularity event information based on a shortargument, the method comprising:
extracting an event short argument and a trigger word corresponding to the event short argument from a text to be extracted by using a short argument extraction model, wherein the event short argument represents an entity argument of which the descriptive information is removed from the event short argument;
extracting the short meta description from the text to be extracted by using a short meta description extraction model; the short argument description represents description information of the event short argument;
and matching and recursing the event shortleaf element, the trigger word corresponding to the event shortleaf element and the shortleaf element description to obtain a fine-granularity event information list.
2. The method for extracting fine-granularity event information based on the shortfalls according to claim 1, wherein the shortfall extraction model is trained by the following steps:
acquiring a sample text and a plurality of sample event arguments in the sample text;
splitting the sample event argument to obtain a sample event short argument, wherein the sample event short argument represents an entity argument directly related to the event;
labeling each sample event short element and sample trigger word in the sample text;
And training the pre-training model by using the marked sample text to obtain the shorthand meta-extraction model.
3. The method for extracting fine-granularity event information based on a shortleaf argument according to claim 2, wherein said splitting the sample event shortleaf argument to obtain the sample event shortleaf argument comprises:
filtering sample event arguments which are irrelevant to the sample trigger words in the sample event arguments;
splitting and filtering remote entity description information in the sample event argument; the remote entity description information represents a position in a text and a existence distance of an entity, but is text information used for describing the entity in semantic information;
and under the condition that the sample event argument is a compound description type entity, splitting and filtering compound description information in the sample event argument to obtain the sample event argument, wherein the compound description type entity represents a text in which entity information and a plurality of description information exist in the form of a combination word.
4. The method for extracting the fine-granularity event information based on the shortnotes according to claim 2, wherein the labeling each of the sample event shortnotes and the sample trigger words in the sample text comprises:
Labeling a sample trigger word, a related event and a short element label corresponding to each sample event short element in the sample text;
and labeling related events and trigger word labels corresponding to the sample trigger words for each sample trigger word in the sample text.
5. The method for extracting fine-granularity event information based on the shortfalls according to claim 1, wherein the shortfall description extraction model is obtained by training the following steps:
acquiring a sample text;
defining a sample shorthand meta description and a sample shorthand meta entity in the sample text, wherein the sample shorthand meta description comprises a composite type description and a long-distance description;
marking the corresponding relation between each pair of the sample short meta description and the sample short meta entity in the sample text;
and training the pre-training model by using the marked sample text to obtain the short argument description extraction model.
6. The method for extracting the event information based on the fine granularity of the shortnotes of claim 1, wherein the matching and recursing the event shortnotes, the trigger words corresponding to the event shortnotes and the shortnotes description comprises:
Filtering, from the short argument descriptions, short argument descriptions that are not related to the event short argument;
according to the event shortcuts, trigger words corresponding to the event shortcuts and the positions of the shortcuts in the text to be extracted, the filtered shortcuts are matched with the event shortcuts one by one, and a plurality of matching results are generated;
and recursion is carried out on the plurality of matching results according to the trigger words corresponding to the event shorthand meta-data to obtain the fine-granularity event information list.
7. The method for extracting fine-granularity event information based on a shorthand meta-data according to claim 6, wherein the recursively obtaining the fine-granularity event information list according to the trigger words corresponding to the event shorthand meta-data includes:
determining a plurality of related candidate matching results from the plurality of matching results according to trigger words corresponding to each event shorthand element;
combining a plurality of candidate matching results to obtain one or more event information of the same event;
and generating the fine-granularity event information list according to the event information of each event.
8. A short argument-based fine granularity event information extraction apparatus, comprising:
the short argument extraction module is used for extracting event short arguments and trigger words corresponding to the event short arguments from the text to be extracted by utilizing a short argument extraction model, wherein the event short arguments represent entity arguments from which description information is removed;
the description extraction module is used for extracting the short meta description from the text to be extracted by using the short meta description extraction model; the short argument description represents description information of the event short argument;
and the event information generation module is used for matching and recursing the event shortleaf element, the trigger word corresponding to the event shortleaf element and the shortleaf element description to obtain a fine-granularity event information list.
9. An electronic device comprising a memory, a processor and a computer program stored on the memory, characterized in that the processor executes the computer program to implement the steps in the method for extracting fine-grained event information based on shortargument as claimed in any one of claims 1 to 7.
10. A computer readable storage medium having stored thereon a computer program/instruction which when executed by a processor implements the steps in the method of extracting fine-grained event information based on arguments as claimed in any one of claims 1 to 7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311352557.7A CN117094397B (en) | 2023-10-19 | 2023-10-19 | Fine granularity event information extraction method, device and product based on shorthand |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311352557.7A CN117094397B (en) | 2023-10-19 | 2023-10-19 | Fine granularity event information extraction method, device and product based on shorthand |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN117094397A true CN117094397A (en) | 2023-11-21 |
| CN117094397B CN117094397B (en) | 2024-02-06 |
Family
ID=88780570
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311352557.7A Active CN117094397B (en) | 2023-10-19 | 2023-10-19 | Fine granularity event information extraction method, device and product based on shorthand |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117094397B (en) |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113705218A (en) * | 2021-09-03 | 2021-11-26 | 四川大学 | Event element gridding extraction method based on character embedding, storage medium and electronic device |
| US20210406295A1 (en) * | 2020-06-30 | 2021-12-30 | Beijing Baidu Netcom Science And Technology Co., Ltd. | Method, electronic device, and storage medium for generating relationship of events |
| CN113946681A (en) * | 2021-12-20 | 2022-01-18 | 军工保密资格审查认证中心 | Text data event extraction method and device, electronic equipment and readable medium |
| CN114036955A (en) * | 2021-10-30 | 2022-02-11 | 西南电子技术研究所（中国电子科技集团公司第十研究所） | Detection method for headword event and argument of central word |
| CN115238685A (en) * | 2022-09-23 | 2022-10-25 | 华南理工大学 | Combined extraction method for building engineering change events based on position perception |
| CN115630304A (en) * | 2022-10-31 | 2023-01-20 | 中国科学技术大学 | Event segmentation and extraction method and system in text extraction task |
| CN115658905A (en) * | 2022-11-07 | 2023-01-31 | 中国电子科技集团公司第二十八研究所 | Cross-chapter multi-dimensional event image generation method |
| CN115965003A (en) * | 2022-12-21 | 2023-04-14 | 中移动信息技术有限公司 | Event information extraction method and event information extraction device |
-
2023
- 2023-10-19 CN CN202311352557.7A patent/CN117094397B/en active Active
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20210406295A1 (en) * | 2020-06-30 | 2021-12-30 | Beijing Baidu Netcom Science And Technology Co., Ltd. | Method, electronic device, and storage medium for generating relationship of events |
| CN113705218A (en) * | 2021-09-03 | 2021-11-26 | 四川大学 | Event element gridding extraction method based on character embedding, storage medium and electronic device |
| CN114036955A (en) * | 2021-10-30 | 2022-02-11 | 西南电子技术研究所（中国电子科技集团公司第十研究所） | Detection method for headword event and argument of central word |
| CN113946681A (en) * | 2021-12-20 | 2022-01-18 | 军工保密资格审查认证中心 | Text data event extraction method and device, electronic equipment and readable medium |
| CN115238685A (en) * | 2022-09-23 | 2022-10-25 | 华南理工大学 | Combined extraction method for building engineering change events based on position perception |
| CN115630304A (en) * | 2022-10-31 | 2023-01-20 | 中国科学技术大学 | Event segmentation and extraction method and system in text extraction task |
| CN115658905A (en) * | 2022-11-07 | 2023-01-31 | 中国电子科技集团公司第二十八研究所 | Cross-chapter multi-dimensional event image generation method |
| CN115965003A (en) * | 2022-12-21 | 2023-04-14 | 中移动信息技术有限公司 | Event information extraction method and event information extraction device |
Non-Patent Citations (3)
| Title |
|---|
| DEBANJANA KAR: "Event Argument Extraction using Causal Knowledge Structures", ARXIV:2105.00477V1 * |
| KANG LIU ETC.: "Extracting Events and Their Relations from Texts: A Survey on Recent Research Progress and Challenges", AI OPEN * |
| 苏杭等: "利用提示调优融合多种信息的低资源事件抽取方法", 计算机应用研究 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN117094397B (en) | 2024-02-06 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107291783B (en) | Semantic matching method and intelligent equipment | |
| CN108345690B (en) | Intelligent question and answer method and system | |
| CN106446232A (en) | Sensitive texts filtering method based on rules | |
| CN102262634B (en) | Automatic questioning and answering method and system | |
| CN104503998B (en) | For the kind identification method and device of user query sentence | |
| CN103309926A (en) | Chinese and English-named entity identification method and system based on conditional random field (CRF) | |
| CN114757176B (en) | Method for acquiring target intention recognition model and intention recognition method | |
| CN106297785A (en) | A kind of intelligent service system based on car networking | |
| CA2610208A1 (en) | Learning facts from semi-structured text | |
| CN108334493B (en) | Question knowledge point automatic extraction method based on neural network | |
| CN113312461A (en) | Intelligent question-answering method, device, equipment and medium based on natural language processing | |
| CN101334933A (en) | Traffic information processing apparatus and method thereof, traffic information integrating apparatus and method | |
| CN103425686B (en) | A kind of information issuing method and device | |
| CN108932278B (en) | Man-machine conversation method and system based on semantic framework | |
| CN104462064A (en) | Method and system for prompting content input in information communication of mobile terminals | |
| CN110309277B (en) | Man-machine conversation semantic analysis method and system, vehicle-mounted man-machine conversation method and system, controller and storage medium | |
| CN105095415A (en) | Method and apparatus for confirming network emotion | |
| CN103678358A (en) | Information search method and system | |
| CN107340766A (en) | Power scheduling alarm signal text based on similarity sorts out and method for diagnosing faults | |
| CN112380868A (en) | Petition-purpose multi-classification device based on event triples and method thereof | |
| CN112527955A (en) | Data processing method and device | |
| CN106156340A (en) | A kind of name entity link method | |
| CN110737770B (en) | Text data sensitivity identification method and device, electronic equipment and storage medium | |
| CN111079428A (en) | Word segmentation and industry dictionary construction method and device and readable storage medium | |
| CN117094397B (en) | Fine granularity event information extraction method, device and product based on shorthand |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |