Information processing method and device
The incremental update method for matter graphs improves efficiency by identifying and updating based on additional events, reducing computational load and enabling real-time updates.
Patent Information
- Application Number
- JP2024043380
- Authority / Receiving Office
- JP · JP
- Patent Type
- Applications
- Current Assignee / Owner
- Priority Date
- 2023-04-26
- Filing Date
- 2024-03-19
- Publication Date
- 2024-11-08
- Estimated Expiration
- 2044-03-19
AI Technical Summary
The construction of matter graphs from increasing volumes of news data is inefficient due to the complexity of the process and the need to rebuild the graph from scratch with each update, leading to low efficiency and increased time requirements as data size grows.
An incremental update method for matter graphs that identifies additional events different from existing events, updates the graph based on these additional events, and utilizes document and key-value databases for efficient storage and correlation calculations.
This approach enhances the efficiency of matter graph construction by reducing computational load and time consumption, allowing for real-time updates and maintaining high temporal sensitivity.
Smart Images

Figure 2024159522000001_ABST
Abstract
Description
[Technical field]
[0001] TECHNICAL FIELD Exemplary embodiments of the present disclosure relate generally to the field of computers, and more particularly to information processing methods and apparatus. [Background technology]
[0002] With the development of network technology and multimedia technology, the number of news is increasing exponentially day by day. Events can be extracted from news and the relationship between different events can be expressed using an event graph. As the number of news increases day by day, the number of events extracted also increases. Therefore, the event graph needs to be updated accordingly. If an event graph is constructed from scratch every time new news is obtained by using the newly obtained news together with past news, the efficiency of constructing the event graph is very low. The construction of an event graph is a complex process. With the increase in the amount of data involved in the construction, the difficulty of constructing the event graph also increases significantly. Therefore, a solution for efficiently constructing an event graph is expected. Summary of the Invention
[0003] In a first aspect of the present disclosure, an information processing method is provided, the method includes: determining a plurality of events to be processed from one or more media contents, where at least two of the plurality of events to be processed have an event relationship, determining an additional event set from the plurality of events to be processed based on an existing event set to construct a matter graph, where the additional event in the additional event set is different from an existing event in the existing event set, and updating the matter graph based on the additional event set.
[0004] In a second aspect of the present disclosure, an electronic device is provided, the electronic device including at least one processing circuit configured to determine a plurality of events to be processed from one or more media contents, at least two of the plurality of events to be processed having an event relationship, determine an additional event set from the plurality of events to be processed based on an existing event set to construct a matter graph, the additional event in the additional event set is different from an existing event in the existing event set, and update the matter graph based on the additional event set.
[0005] In some embodiments of the second aspect, the at least one processing circuit is further configured to determine, for a given event among the plurality of events to be processed, a correspondence match between the text describing the given event and the text describing each existing event in the set of existing events, and in response to all of the correspondence matches being less than a threshold match, identify the given event as the additional event. In some embodiments of the second aspect, the at least one processing circuit is further configured to add an item corresponding to the given event to a document-based database for the event graph to store information related to the given event, and set an added event field in the item to a predetermined value indicating the added event.
[0006] In some implementations of the second aspect, the at least one processing circuit is further configured to cancel identifying the given event as the additional event in response to updating the event graph.
[0007] In some embodiments of the second aspect, the at least one processing circuit is further configured to, in response to a match between the text describing the given event and the text describing the first existing event exceeding a threshold match, store information related to the target media content in the one or more media contents in the document-based database and in an item corresponding to the first existing event, and the given event is determined from the target media content.
[0008] In some embodiments of the second aspect, the at least one processing circuit further determines, for a first additional event in the set of additional events, a first text element in a text describing the first additional event, updates element occurrence frequency information stored in the key-value database and corresponding to the first text element, determines a second existing event in the set of existing events, where the text describing the second existing event includes the first text element, and updates a correlation degree of the event relationship between the second existing event and a third existing event in the event graph based on the updated element occurrence frequency information.
[0009] In some embodiments of the second aspect, the at least one processing circuit is further configured to determine a first event and a second text element in the text describing the first event that have an event relationship with the first additional event in the existing event set or the additional event set, update element-pair occurrence frequency information stored in the key-value database indicating that the first text element occurs in relation to the second text element, and in response to the text describing a third existing event including the second text element, further update the correlation degree based on the updated element-pair occurrence frequency information.
[0010] In some embodiments of the second aspect, the at least one processing circuitry further determines, for a second additional event in the set of additional events, a similarity between the second additional event and a second event in the set of additional events or the set of existing events, and in response to the similarity exceeding a threshold similarity, stores an indication of the similarity relationship between the second additional event and the second event in the document-based database for the matter graph.
[0011] In some embodiments of the second aspect, the at least one processing circuit is further configured to summarize the second additional event and an abstract event for at least the second event having a similarity relationship with the second additional event, map the second additional event and the second event to the abstract event in a matter graph, and add a node representing the abstract event to a visual representation of the matter graph.
[0012] In a third aspect of the present disclosure, there is provided an electronic device, the device including at least one processing unit and at least one memory, the at least one memory coupled to the at least one processing unit and configured to store instructions for execution by the at least one processing unit, the instructions when executed by the at least one processing unit causing the device to perform the method of the first aspect.
[0013] In a fourth aspect of the present disclosure, there is provided a computer readable storage medium having stored thereon a computer program which, when executed by a processor, implements the method of the first aspect.
[0014] It should be noted that the contents described in the Summary of the Invention are not intended to limit the key or important features of the embodiments of the present disclosure, nor are they intended to limit the scope of the present disclosure. It should be understood that other features of the present disclosure can be more easily understood by reading the following description. [Brief description of the drawings]
[0015] The above and other features, advantages and aspects of the embodiments of the present disclosure will become more apparent from the following detailed description taken in conjunction with the accompanying drawings, in which the same or similar reference numerals indicate the same or similar elements.
[0016] [Figure 1] FIG. 1 is a schematic diagram illustrating an example environment in which embodiments of the present disclosure may be implemented. [Diagram 2] FIG. 1 is a schematic diagram illustrating an example architecture of an information processing system according to some embodiments of the present disclosure. [Diagram 3] FIG. 2 is a schematic diagram illustrating an event processing flow according to some embodiments of the present disclosure. [Figure 4] FIG. 2 is a schematic diagram illustrating the runtime of incremental construction of an event graph according to some embodiments of the present disclosure. [Diagram 5] 1 is a flow chart illustrating a process for information processing according to some embodiments of the present disclosure. [Figure 6]FIG. 1 is a block diagram illustrating a number of example electronic devices embodying the present disclosure. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS
[0017] Hereinafter, the embodiments of the present disclosure will be described in more detail with reference to the accompanying drawings. Although the accompanying drawings show some embodiments of the present disclosure, it should be understood that the present disclosure can be implemented in various forms and is not limited to the embodiments herein, and these embodiments are used to more thoroughly and completely understand the present disclosure. It should be understood that the accompanying drawings and embodiments of the present disclosure are used for illustrative purposes only and are not intended to limit the scope of protection of the present disclosure.
[0018] It should be noted that any section / subsection titles provided herein are not limiting. Various embodiments are described throughout this specification, and any embodiment may be included in any section / subsection. Furthermore, any embodiment described in any section / subsection may be combined in any manner with any other embodiment described in the same section / subsection and / or in a different section / subsection.
[0019] In describing embodiments of the present disclosure, the term "comprising" and similar terms should be understood in their broad inclusion, i.e., "including, but not limited to." The term "based on" means "based at least in part on." "In one embodiment" or "the embodiment" means "at least one embodiment." "Some embodiments" means "at least some embodiments." The following may include explicit and implicit definitions.
[0020] The term "circuitry" as used herein means hardware circuitry and / or a combination of hardware circuitry and software. For example, a circuitry may be a combination of analog and / or digital hardware circuitry and software / firmware. As another example, a circuitry may be any portion of a hardware processor including software, where the hardware processor includes digital signal processor(s), software, and memory(s) that work together to operate the device and perform various functions. In another example, a circuitry may be a hardware circuitry and / or processor, such as a microprocessor or a portion of a microprocessor, that requires software / firmware for operation, but may be absent if not required to operate. The term "circuitry" as used herein may include implementation of only a hardware circuitry or processor, or a portion of a hardware circuitry or processor and its (or their) associated software and / or firmware.
[0021] The term "event" as used herein refers to the occurrence of some action or situation in which participants participate, or a change in an objective state. A text describing an event may include multiple words and is used to describe the occurrence of an event or some constituent part of an event. Formally, the causes of an event may include trigger words or type of event, main participants in the event, time and place of occurrence of the event, etc.
[0022] The term "incremental update" as used herein means that during an update, only the changed data is updated, and the unchanged or updated data is not updated repeatedly, thereby saving the update operation time and improving efficiency.
[0023] The term "text" as used herein means any length of language. By way of example, text may mean one or more words, phrases, parts of a sentence, sentences, etc.
[0024] The term "similar event pair" refers to events that have different descriptive text but express the same meaning. For example, the events "price rise" and "price increase" are a similar event pair.
[0025] The term "word" as used herein may have any suitable granularity. For example, in one language, a "word" may include one or more words. In another language, a "word" may be a single word made up of one or more characters.
[0026] Example Environment 1 is a schematic diagram of an example environment 100 in which embodiments of the present disclosure may be implemented. In the environment 100, an electronic device 120 obtains one or more media content 110-1, 110-2, ..., 110-N, collectively or singly referred to as media content 110, where N is an integer equal to or greater than 1. The electronic device 120 may construct an event graph 130 based on the media content 110.
[0027] In environment 100, electronic device 120 may be any device having computing capabilities, including a terminal device. A terminal device may be any mobile, fixed or portable terminal, including a mobile phone, a desktop computer, a laptop computer, a notebook computer, a netbook computer, a tablet computer, a media computer, a multimedia computer, a personal communication system (PCS) device, a personal navigation device, a personal digital assistant (PDA), an audio / video player, a digital camera / camera, a locator device, a television receiver, a radio broadcast receiver, an e-book device, a gaming device, or any combination thereof, accessories and peripherals including such devices, or any combination thereof.
[0028] The media content 110 may be any suitable content capable of providing information. For example, the media content 110 may be a news report in a format such as text, image, audio, video, or a combination thereof. The media content 110 may be a news report in any sector, industry, or field, such as news in the financial sector. The media content 110 may be media content obtained from various platforms (e.g., news platforms) or may be stored media content. For media content 120 in text format, the electronic device 120 may directly extract information for constructing the event graph 130 from the text. For media content 110 in the format of images, videos, audio, etc., the electronic device 120 may use any existing or future developed technology to extract information for constructing the event graph 130 from the images, audio, or video. For example, the electronic device 120 may directly extract relevant information from image, video, or audio formats based on image identification or voice identification technology.
[0029] An event graph is an event-logic knowledge base that describes evolutionary rules and patterns among events. Event graphs are used to represent events and event relationships among different events. For example, an event graph can represent events and event relationships using a logical directed graph. Such a logical directed graph has events as nodes and event relationships as directed edges. Constructing an event graph is a complex process. For example, the construction process may include steps such as event extraction, relationship extraction, event generalization, relationship strength calculation, and knowledge storage.
[0030] In this specification, an event relationship is also referred to as an association relationship or simply a relationship. Such event relationships may include causal relationships, conditional relationships, inversion relationships, order relationships, superior-subordinate relationships, compositional relationships, simultaneous relationships, similarity relationships, and the like. As an example, a causal relationship is a relationship in which the occurrence of a previous event (cause) leads to the occurrence of a later event (result). A conditional relationship refers to a previous event being a condition for the occurrence of the later event. An inversion relationship refers to a relationship in which one event is opposed to another event, for example, one event occurs late while another event occurs early. A sequential relationship refers to the occurrence of a previous event and a later event successively in time. A superior-subordinate relationship refers to an event being a superior or subordinate event of another event, and includes noun superior-subordinate and verb superior-subordinate. For example, the event "food price rise" and the event "vegetable price rise" constitute a noun superior-subordinate relationship, and the event "murder" and the event "assassination" constitute a verb superior-subordinate relationship. A compositional relationship refers to an event being a compositional part of another event. A simultaneous relationship refers to an event occurring simultaneously with another event. A similarity relationship refers to an event being similar to another event to a certain extent, and establishing the similarity relationship through, for example, similarity calculation. The above event relationships are merely examples and are not intended to limit the scope of the present disclosure.
[0031] It should be understood that the structure and functionality of environment 100 is described for illustrative purposes only and is not meant to limit the scope of the present disclosure.
[0032] As briefly mentioned above, constructing an event graph is a process with complicated steps and many intermediate results. In some scenarios, the event graph has a high time frame requirement. For example, a financial event graph based on causal relationships can show the causal transmission chain between financial events and can be used to predict financial risk events in advance. Time sensitivity is very important in predicting financial risk events. When using an event graph to make financial risk predictions, only by incorporating time-sensitive financial events into the event graph in a timely manner can risk events be predicted more reasonably and accurately. Therefore, constructing a time-sensitive event graph is of great significance.
[0033] On the other hand, the amount of media content, such as news, is growing rapidly. If the process of constructing the graph cannot be done incrementally, then every time the graph is updated, we have to start from scratch to reconstruct the event graph using the new information together with the past information. As the graph scale increases, the graph update time also becomes longer and longer. This may lead to a very low construction efficiency of the event graph.
[0034] Considering this, it is necessary to update the event graph incrementally. How to update the event graph incrementally and efficiently is a non-trivial issue.
[0035] Therefore, an embodiment of the present disclosure provides a solution for information processing, in which a plurality of events to be processed are determined from one or more media contents, and at least two of the events to be processed have an event relationship; a set of additional events is determined from the plurality of events to be processed based on a set of existing events for constructing a matter graph; the additional events in the set of additional events are different from the existing events in the set of existing events; i.e., the existing events are filtered from the events to be processed, and the additional events not considered in the current matter graph are screened; and the matter graph is updated based on the set of additional events.
[0036] Therefore, the event graph is updated based on the additional events that are different from the existing events. In this way, the event graph can be updated only based on the additional events, and the event graph can be updated incrementally. This improves the efficiency of constructing the event graph.
[0037] Matter graph update architecture 2 is a schematic diagram of an exemplary architecture 200 of information processing according to some embodiments of the present disclosure. The architecture 200 includes a media content acquisition module 211, a pre-processing module 212, an event extraction module 213, an event identification module 214, a correlation degree incremental calculation module 215, a similarity degree incremental module 216, an event generalization module 217, a correlation degree calculation module 218, and an event graph storage module 219. The architecture 200 may further include a data storage structure 220, a pre-processing data storage base 221, a related event storage base 222, a document-based database 223, a key-value-based database 224, an abstract event pair storage base 225, and an event graph storage base 226. These modules in the architecture 200 may be implemented in the electronic device 120.
[0038] In some embodiments, a plurality of events (also referred to as a plurality of events to be processed) may be determined from one or more media contents 110. At least two of the events to be processed have an event relationship, i.e., an associative relationship. An event pair having an associative relationship may be a combination of two events that have a logical relationship, such as sequential, causal, conditional, superior-subordinate, etc.
[0039] The following describes examples of events to be determined and processed in relation to the media content acquisition module 211, the pre-processing module 212, and the event extraction module 213 in the architecture 200.
[0040] The media content acquisition module 211 is configured to acquire one or more media content 110. For example, the media content acquisition module 211 may acquire media content 110 from various platforms (e.g., news platforms) or may acquire stored media content 110 from any storage medium.
[0041] In some implementations, the media content acquisition module 211 may use crawler techniques to acquire media content 110 from the platform. The media content acquisition module 211 may acquire media content from the platform periodically or at any time to maintain the time sensitivity of the media content (e.g., news data), and may acquire media content from the platform in real time.
[0042] The preprocessing module 212 is configured to preprocess the media content 110 acquired by the media content acquisition module 211. For example, preprocessing of news prereading may include, but is not limited to, segmentation, clause division, special punctuation processing, word segmentation, vocabulary labeling or other preprocessing methods. After the preprocessing is completed, the preprocessing module 212 may store the results obtained by preprocessing the media content 110 in the preprocessing data storage base 221.
[0043] The event extraction module 213 is configured to extract events from the pre-processed media content. The extracted events are events to be processed. For example, event pairs having an associative relationship may be extracted from the pre-processed media content. The event extraction module 213 may extract event pairs having a logical relationship, such as order, causality, condition, and / or superiority / subordinateity, from the pre-processed media content.
[0044] In some embodiments, the event extraction module 213 may use a deep learning model to extract event pairs having an associative relationship from the pre-processed media content. The deep learning model may be trained using a corpus labeled with event pairs having an associative relationship. Any suitable algorithmic deep learning model may be employed. For example, a BERT model, a Bidirectional Long Short-Term Memory Network (BiLSTM), or a Conditional Random Field (CRF) model may be employed. As another example, a BERT_BiLSTM_CRF model may be employed.
[0045] In some embodiments, the event extraction module 213 may store the extracted event pairs with the association relationships in the related event storage base 222 .
[0046] An example of an event to be determined will be described with reference to FIG. 3. FIG. 3 is a schematic diagram of an event processing flow 300 according to some embodiments of the present disclosure. As shown in FIG. 3, the media content 301 is "The rise in crude oil prices leads to an increase in the company's road transport business costs, and the company promptly formulates a corresponding pricing strategy in consideration of market conditions, etc." Through pre-processing and event extraction of the media content 301, a pair of events having an association relationship, namely, a "crude oil price rise" event 311 and a "rise in the company's road transport business costs" event 312, can be extracted from the media content 301. Since the "crude oil price rise" is the cause of the result "rise in the company's road transport business costs", here, the two events "crude oil price rise" and "rise in the company's road transport business costs" are an event pair having an association relationship, specifically, they are an event pair having a causal relationship.
[0047] Continuing with reference to Figure 2, the event identification module 214 determines a set of additional events from the plurality of events to be processed based on the set of existing events that construct the incident graph 130. The additional events in the set of additional events are different from the existing events in the set of existing events and are also referred to as additional events.
[0048] The existing event set refers to a set of existing events that have already been used to construct the matter graph. That is, the existing events are already embodied in the matter graph. For example, the existing events may be events that were previously extracted from media content to construct the matter graph. In contrast, the additional events refer to events that have not yet been used to construct the matter graph, for example events extracted from newly acquired media content.
[0049] The event identification module 214 may use any suitable method to compare the event to be processed with the existing events to identify additional events in the event to be processed. In some embodiments, for any event to be processed (also referred to as a given event) in the plurality of events to be processed, the event identification module 214 may determine a correspondence match between the text describing the given event and the text describing each existing event in the set of existing events. If the correspondence matches are all below a threshold match, the given event may be identified as an additional event.
[0050] In some embodiments, the degree of match is determined by comparing character strings contained in the text describing the given event and the text describing each existing event. For example, if the number of identical characters contained in the text describing the given event and the text describing an existing event in the existing event set is equal to or greater than a threshold number (or if both contain exactly the same character string), the degree of match between the text of the given event and the text of the existing event is determined to be high. This means that the given event is not an additional event for the event graph. If the number of identical characters contained between the text describing the given event and the text describing each existing event in the existing event set is all less than a threshold number (or if the given event and any existing event do not contain exactly the same character), the degree of match between the text of the given event and the text of all existing events in the existing event set is determined to be low. This means that the given event is an additional event for the event graph.
[0051] In some embodiments, the event identification module 214 may assign a unique identifier (ID) to each extracted event (i.e., each event to be processed). The ID may be a unique identifier for each event (including the event to be processed and the existing event). The event identification module 214 may store information of the additional event in a document-based database for event graphs 223. The document-based database 223 may include, but is not limited to, a MongoDB database or other document-based databases.
[0052] The document-based database 223 includes items for storing information about events. Each item is used to store information about a corresponding event, and the assigned event ID may be used as an index of the item. The item may include a field for identifying whether the corresponding event is an additional event, also called an additional event field. During the process of constructing the event graph, since there are many attributes of the event and the related data format is complicated, the present disclosure adopts the document-based database 223 as a data storage manner, which is convenient for adding, deleting, modifying, and checking data and contributes to the gradual construction of the event graph.
[0053] In some embodiments, if a given event is identified as an add event, the event identification module 214 may add an item corresponding to the given event to the document database 223. The event identification module 214 may set an add event field in the item to a predefined value to indicate that the given event is an add event. As an example, the field "new" may be added to the item to identify the corresponding event as an add event.
[0054] After each round or batch of updates to the event graph, the event identification module 214 may cancel the identification of the added event. For example, the event identification module 214 may modify the added event field of the item of the added event in that round of updates to a value indicating it is an existing event, or may remove the added event field from the item.
[0055] In some embodiments, if the degree of match between the text describing the given event and the text describing an existing event (also called a first existing event) in the set of existing events exceeds a threshold degree of match, it means that the given event is not an additional event but is the existing event. In this case, information related to the target media content may be stored in the item corresponding to the existing event in the document-type database 223. The given event is determined from the target media content. In other words, the media content used in extracting the given event may be stored in the item corresponding to the existing event as another source of the existing event.
[0056] After the additional events are identified, the matter graph is then updated based on the additional event set, thereby realizing the incremental update of the matter graph. Through the incremental update of the matter graph, the original static matter graph can be transformed into a dynamically updated matter graph, and the media content can be constructed in the matter graph in real time.
[0057] The event graph includes various information such as events, association relationships between different events, correlation degrees of the association relationships, etc. Incremental updating of the event graph may include incremental updating of one or more pieces of information.
[0058] The correlation degree of the association relationship indicates how likely or how likely a pair of associated events have that association relationship. For example, the causal strength of the causal relationship indicates how likely a cause event causes an effect event. In some embodiments, the correlation degree may be updated incrementally. Thus, as shown in FIG. 2, the architecture 200 may further include a correlation degree incremental calculation module 215 configured to update the correlation degree of the association relationship between the events using additional events in the additional event set.
[0059] The correlation degree can be calculated by any suitable algorithm. In some embodiments, the correlation degree may be calculated based on the frequency of occurrence of words (also called word frequency) and the frequency of occurrence of word pairs (also called word pair co-frequency) in the text describing the event. Hereinafter, the calculation of the correlation degree will be described by taking a causal relationship as an example. In the case of a causal relationship, the correlation degree is also called causal strength.
[0060] As an example, an event A may cause another event B to occur. Thus, event A is called the cause event, and event B is called the effect event. In a causal chain, causal strength reflects the probability that a cause event leads to the occurrence of an effect event. For example, in a causal event pair (A,B), necessary causation means that cause event A must be present for effect event B to occur, and sufficiency causation means that all cause events A are necessary to bring about effect event B. In calculating causal strength, we first consider the words in cause event A,
number
number
[0061] An example of a word pair and event pair causal strength calculation will be described below. First, a word pair is calculated from the viewpoint of necessity and sufficiency.
number
number
[0062] Where:
number
number
number
number
number
number
[0063] Here, the word
number
number
number
number
number
number
number
number
[0064] Where:
number
number
number
number
[0065] As a result, the word pair
number
number
number
[0066] Where:
number
number
[0067] TIFF2024159522000029.tif11153Here, indicates the causal strength of events A and B.
[0068] The above describes the determination of correlation degree by taking causal strength as an example. In the calculation of correlation degree, a relatively large number of intermediate results are generated, such as frequency information of cause words, frequency information of result words, frequency information of related word pairs, necessary causal fraction of related word pairs, sufficient causal fraction of related word pairs, and causal fraction of result word pairs, etc. These intermediate results need to be read and written frequently.
[0069] Therefore, in some embodiments, the key-value database 224 can be used to store intermediate results of the correlation calculation. The key-value database 224 can be, for example, a Redis cache database or other key-value database. By storing data that needs to be frequently read and written in a Redis cache database or the like, the efficiency of reading and writing data can be improved, thereby improving the efficiency of building the event graph. Furthermore, the problem of memory full that may exist in the process of building the event graph can be easily solved.
[0070] In such an embodiment, the correlation degree incremental calculation module 215 incrementally updates the correlation degree in the event graph. For any additional event (also referred to as a first additional event) in the additional event set, the correlation degree incremental calculation module 215 can determine each text element (also referred to as a first text element, e.g., a word) in the text describing the additional event, and correspondingly update the element occurrence frequency information stored in the key-value database 224 and corresponding to the text element. The correlation degree incremental calculation module 215 can further determine an existing event (also referred to as a second existing event) that includes the text element in the existing event set, i.e., the text describing the existing event includes the text element. The correlation degree incremental calculation module 215 can further update the correlation degree of the event relationship between the existing event and one or more other existing events (also referred to as a third existing event) in the event graph based on the updated element occurrence frequency information of the text element.
[0071] As an example, an additional event "Heavy Rain at Location A" is identified in the update of the current round, which means that the frequency of occurrence of the word "heavy rain" increases. Therefore, the frequency information of occurrence of the word "heavy rain" is updated in the key-value database. If there is an existing event "Heavy Rain at Location B", the word "heavy rain" is also included. Since the frequency of occurrence of the word "heavy rain" has changed, the correlation degree between the existing event "Heavy Rain at Location B" and other existing events that have a related relationship with it is also updated.
[0072] In some embodiments, the appearance of the additional event may also cause the updating of element pair occurrence frequency information (e.g., word pair occurrence frequency information). Thus, the correlation degree may be further updated based on the updated element pair occurrence frequency information. Specifically, the correlation degree incremental calculation module 215 may determine an event (also called a first event or an associated event) having an association relationship with an additional event (e.g., the above-mentioned first additional event) in the existing event set or the additional event set, and a text element (also called a second text element) in the text describing the associated event. This means that the occurrence frequency of an element pair consisting of the first text element and the second text element increases. Thus, the correlation degree incremental calculation module 215h updates the element pair occurrence frequency information, e.g., word pair occurrence frequency information, stored in the key-value database 224, in which the first text element occurs in relation to the second text element. If the text describing the third existing event also includes the second text element, the correlation degree between the second existing event and the third existing event is further updated based on the updated element pair occurrence frequency information.
[0073] As an example, assume that the first added event in the current round of update is "Heavy Rain at Location A" and the first text element is "Heavy Rain". The related event related to the first added event is "Flood at Location A", and the second text element in the text of the first event is "Flood". Next, it is necessary to update the occurrence frequency information of the element pair "Heavy Rain"-"Flood". If the third existing event that includes "Flood" is "Flood at Location B", the correlation degree between the second existing event "Heavy Rain at Location B" and the third existing event "Flood at Location B" is further updated based on the occurrence frequency information of the element pair "Heavy Rain"-"Flood".
[0074] In this embodiment, in each round of updating the event graph, search for additional events based only on the additional event indicators, and use these additional events to update the word frequency and causal word pair frequency. At the same time, select a key-value database to store frequency information, and store the frequency information of text elements and the frequency information of text element pairs in the key-value database. This allows efficient reading and writing when updating frequency information, and greatly improves the efficiency of calculating correlation.
[0075] The above describes the incremental calculation of correlation. In some embodiments, the similarity between events may be calculated incrementally. As shown in FIG. 2, the architecture 200 may further include a similarity incremental calculation module 216 configured to determine whether an additional event in the additional event set has a similar event. For any additional event (also referred to as a second additional event) in the additional event set, the similarity incremental calculation module 216 may determine a similarity between the second additional event and any event (also referred to as a second event) in the additional event set or the existing event set. If the similarity exceeds a threshold similarity, an indication of the similarity relationship between the second additional event and the second event is stored in the document database 223.
[0076] When the similarity between the second additional event and the second event exceeds the threshold similarity, it means that the similarity between the second additional event and the second event is high, so the second additional event and the second event may be called a similar event pair. Information of the similar event pair may be stored in the document-type database 223. For example, if event 1 includes three similar events, event 2, event 3, and event 4, an indication of the similarity relationship between these four events, i.e., the similarity relationship between the events, may be stored. An exemplary storage format may be {event_id:1,sim_events:[2,3,4]}, which is used to indicate that the similar events of event 1 include event 2, event 3, and event 4.
[0077] In the embodiments of the present disclosure, any suitable similarity calculation method can be used. For example, Jaccard similarity, Pearson correlation coefficient, Euclidean distance similarity, cosine similarity, etc. can be adopted. Of course, other methods that can be used for similarity calculation can also be adopted.
[0078] In each round of updating the matter graph, the additional events can be determined by identifying the additional events. Assume that the number of existing events in the matter graph is a, and the number of additional events in the current round is b. Before the current round, the similarity between any two existing events has already been calculated. In the update of the current round, only the similarity between two events among the b additional events and the similarity between two combinations of the a existing events and the b additional events need to be calculated. Therefore, the number of similarity calculations required for the matter graph in the update of the current round is
number
number
number
number
[0079] In some embodiments, the similarity calculations required for the current round of event graph updates may use multi-process or multi-thread parallelism to further increase the efficiency of event graph updates.
[0080] The event generalization module 217 is configured to perform matter logical abstraction of the event to obtain an abstract event. For example, if an event has two or more similar events, its occurrence is considered to have a general nature and can be generalized. It should be noted that, compared with the abstract event, the event extracted from the media content is also called a concrete event.
[0081] The event generalization module 217 may perform generalization based on the calculation result of the event similarity. In each round of incremental update, when a similar event is added to an event, the abstract event generalized by the event and its similar event may be updated. For example, for the above-mentioned second added event and its one or more similar events (also referred to as second events), the event generalization module 217 may determine an abstract event used to summarize the second added event and its similar events. Furthermore, the event generalization module 217 may correspond (e.g., map) the second added event and its similar events to the abstract event in the matter graph. The event generalization module 217 may further add a node indicating the abstract event to a visual representation of the matter graph.
[0082] By way of example and without intending to be limiting, an exemplary process of event generalization is described. For example, the text describing these similar events can be first subjected to word segmentation and word property marking, and then meaningless words can be removed according to the word property. For example, a word whose word property is an exclamation can be designated as a meaningless word. Then, a generalized event can be obtained by extracting and combining common words in the text. If there is only one common word, determine whether a core word in the text describing the event is an object. If the core word is the common word and the core word is not an object, the common word can be used as a generalized event, i.e., the common word can be used as the text describing the abstract event. The core word here can be a word that captures a part of the meaning in the event.
[0083] The correlation calculation module 218 is configured to calculate the correlation between abstract events. The calculation of the correlation between abstract events is similar to the calculation of the correlation between two events A and B described above, so it will not be repeated here.
[0084] In some embodiments, the generalization results obtained by the event generalization module 217 and the correlation results of the abstract events obtained by the correlation calculation module 218 may be stored in the abstract event pair storage base 225 .
[0085] The event graph storage module 219 is configured to store the aforementioned calculation results in the form of nodes, edges, and attributes in the event graph storage base 226 to form a visualization graph. For example, events may be stored in the event graph storage base 226 in the form of nodes, association relationships may be stored in the event graph storage base 226 in the form of edges, and correlation degrees may be stored in the event graph storage base 226 in the form of attributes of edges. In some embodiments, the event graph storage base 226 may be a graph database, such as the Neo4j graph database, or other graph databases. The embodiments of the present disclosure are not limited in this respect.
[0086] Above, an exemplary architecture for incremental event graph updates has been described with reference to Fig. 2. In the following, an event processing flow 300 will be described with reference to a specific example in Fig. 3.
[0087] 3, the event extraction module 213 can extract an event 311 and an event 312 having a causal relationship from the media content 301. In this example, the event 311 is a cause event "oil price rise", and the event 312 is an effect event "company's road transport business cost rise". The correlation degree progressive calculation module 215 calculates the correlation degree between the event 311 and the event 312, and calculates that the correlation degree between the event 311 and the event 312 is 0.25.
[0088] The similarity progressive calculation module 216 performs event similarity calculation for the event 311 and the event 312, respectively. It can determine similar events 321 for the event 311 and similar events 322 for the event 312. In this example, the similar events 321 include names such as "natural gas price rise" and "wheat price rise", and the similar events 322 include names such as "cost of living rise" and "flour production cost rise", etc.
[0089] Event generalization module 217 performs event generalization on event 311 and its similar events 321 to determine abstract event 331. Event generalization module 217 performs event generalization on event 312 and its similar events 322 to determine abstract event 332. In this example, abstract event 331 is "price increase" and abstract event 332 is "cost increase."
[0090] The correlation calculation module 218 calculates the abstract event correlation between the abstract event 331 and the abstract event 332. Thus, the correlation between the abstract event 331 and the abstract event 332 is calculated to be 0.55. Thus, the event graph is gradually updated. It should be understood that the specific number of events and correlations shown in FIG. 3 are merely examples and do not limit the scope of the present disclosure.
[0091] The solution of the present disclosure can achieve not only efficient incremental construction of the event graph by incrementally updating the event graph to add additional events to the original event graph, but also real-time updating of the event graph.
[0092] The incremental update solution described above may be used to build other types of graphs. In some embodiments, a knowledge graph may be incrementally updated, where each node in the knowledge graph represents an entity. In this case, the above embodiment describing an "event" may be applied to the "entity". For example, one or more additional entities or additional entities that are different from the existing entities in the knowledge graph may be determined from the newly obtained or newly extracted entities, and then the knowledge graph may be updated with these additional entities. In some embodiments, a causal graph may be incrementally updated, where each node in the causal graph represents an element (e.g., a cause element or an effect element). In this case, the above embodiment describing an "event" may be applied to the "element". For example, one or more additional elements or additional elements that are different from the existing elements in the causal graph may be determined from the newly obtained or newly extracted elements, and then the causal graph may be updated with these additional elements.
[0093] Table 1 shows a comparison of the results of using the incremental construction method of the present solution and the results of not using the incremental construction method. It should be understood that Table 1 is explained using a causal event graph having causal relationships as an example. [Table 1]
[0094] As can be seen from Table 1, from the first day (1d) to the twelfth day (12d), a series of indicators indicating the construction results of the event graph all show the construction results of matching, such as concrete event matching rate, causal event pair matching rate, similar event pair matching rate, abstract event matching rate, abstract causal event pair matching rate, and concrete-abstract event projection relationship matching rate. As can be seen from this, the event graph constructed using the incremental construction method of the present disclosure is completely the same as the result generated without using the incremental construction method, which means that the incremental construction method proposed by the present disclosure is correct.
[0095] Table 2 shows the time it takes to build an event graph using the incremental construction method of this disclosure. [Table 2]
[0096] Table 3 shows the time it takes to build an event graph without using augmentation construction. [Table 3]
[0097] Tables 2 and 3 take the causal event graph with causal relationships as an example. As can be seen from the comparison of Tables 2 and 3, the adoption of the solution of the present disclosure greatly reduces the consumption of event similarity calculation and total time over time, and improves the efficiency of event graph construction. As can be seen, with the incremental improvement, the efficiency advantage of incremental construction becomes more and more obvious as the graph scale increases.
[0098] 4 is a schematic diagram of the runtime of the incremental construction of an event graph according to some embodiments of the present disclosure. As shown in FIG. 4, the cumulative time of constructing the event graph changes linearly with the number of accumulated media contents. The event graph needs to be reconstructed for the media contents acquired in each round, and the cumulative time of constructing the event graph changes exponentially with the increase in the number of media contents. As can be seen, when updating the event graph using the same amount of samples, the incremental construction method does not increase the time consumed with the increase in the existing event graph, and the efficiency of updating the event graph is greatly improved.
[0099] Example process 5 is a flow chart of a process 500 for processing information according to some embodiments of the present disclosure. Process 500 may be performed by electronic device 120. For ease of discussion, process 500 will be described with reference to FIG.
[0100] In block 510, the electronic device 120 determines a plurality of events to be processed from the one or more media contents, where at least two of the events to be processed have an event relationship.
[0101] In block 520, the electronic device 120 determines an additional event set from the plurality of events to be processed based on the existing event set for constructing the event graph, where the additional event in the additional event set is different from the existing event in the existing event set.
[0102] In some embodiments, to determine the set of additional events, the electronic device 120 determines, for a given event in the plurality of events to be processed, a correspondence match between text describing the given event and text describing each existing event in the set of existing events, and identifies the given event as an additional event in response to all of the correspondence matches being less than a threshold match.
[0103] In some embodiments, to identify a given event as an additional event, the electronic device 120 adds an item corresponding to the given event to a document-based database for the event graph, stores information related to the given event, and sets an additional event field in the item to a predetermined value indicating the additional event.
[0104] In some embodiments, process 500 may further include electronic device 120 storing information related to the target media content in the one or more media content items in the document-based database and in an item corresponding to the first pre-existing event in response to a match between the text describing the given event and the text describing the first pre-existing event exceeding a threshold match, wherein the given event is determined from the target media content.
[0105] In block 530, the electronic device 120 updates the event graph based on the additional event set.
[0106] In some embodiments, the process 500 may further include the electronic device 120 canceling the identification of the given event as an additional event in response to updating the event graph.
[0107] In some embodiments, to update the event graph, the electronic device 120 may further include: for a first additional event in the set of additional events, determining a first text element in a text describing the first additional event, updating element occurrence frequency information stored in the key-value database and corresponding to the first text element, determining a second existing event in the set of existing events, where the text describing the second existing event includes the first text element, and updating a correlation degree of an event relationship between the second existing event and a third existing event in the event graph based on the updated element occurrence frequency information.
[0108] In some embodiments, to update the correlation degree, the electronic device 120 determines a first event and a second text element in the text describing the first event that have an event relationship with a first additional event in the existing event set or the additional event set, updates element-pair occurrence frequency information stored in the key-value database indicating that the first text element occurs in association with the second text element, and in response to the text describing a third existing event including the second text element, further updates the correlation degree based on the updated element-pair occurrence frequency information.
[0109] In some embodiments, the process 500 may further include the electronic device 120 determining, for a second additional event in the set of additional events, a similarity between the second additional event and the set of additional events or a second event in the set of existing events, and, in response to the similarity exceeding a threshold similarity, storing an indication of the similarity relationship between the second additional event and the second event in the document-based database for the matter graph.
[0110] In some embodiments, to update the event graph, the electronic device 120 determines an abstract event to summarize the second additional event and at least a second event having a similarity relationship with the second additional event, maps the second additional event and the second event to the abstract event in the event graph, and adds a node representing the abstract event to the visual representation of the event graph.
[0111] Example Device 6 is a block diagram of one or more example electronic devices 600 in which the present disclosure may be implemented. It should be understood that the electronic device 600 shown in FIG. 6 is merely an example and is not intended to limit in any way the functionality and scope of the examples described herein. The electronic device 600 shown in FIG. 6 may be used to implement the electronic device 120 of FIG. 1.
[0112] As shown in Figure 6, the electronic device 600 is in the form of a general purpose electronic device. The components of the electronic device 600 include, but are not limited to, one or more processors or processing units 610, memory 620, storage device 630, one or more communication units 640, one or more input devices 650 and one or more output devices 660. The processing units 610 may be real or virtual processors and perform various processes according to programs stored in the memory 620. In a multiprocessor system, multiple processing units execute computer-executable instructions in parallel to increase the parallel processing capabilities of the electronic device 600.
[0113] Electronic device 600 typically includes a number of computer storage media. Such media may be any available media accessible by electronic device 600, including, but not limited to, volatile and non-volatile media, removable and non-removable media. Memory 620 may be volatile memory (e.g., registers, cache, random access memory (RAM)), non-volatile memory (e.g., read only memory (ROM), electrically erasable programmable read only memory (EEPROM), flash memory), or a combination thereof. Storage device 630 may be removable or non-removable media, machine-readable media, such as flash memory drives, magnetic disks, or any other media, for storing information and / or data (e.g., training data for training) and accessible within electronic device 600.
[0114] The electronic device 600 may further include other removable / non-removable, volatile / non-volatile storage media. Although not shown in FIG. 6, a magnetic disk drive capable of reading from or writing to a removable non-volatile magnetic disk (e.g., a "floppy disk") and an optical disk drive capable of reading from or writing to a removable non-volatile optical disk may be provided. In these configurations, each drive is connected to a bus (not shown) via one or more data medium interfaces. The memory 620 may include a computer program product 625 having one or more program modules, which are configured to perform various methods or operations of various embodiments of the present disclosure.
[0115] The communication unit 640 can communicate with other electronic devices via a communication medium. Furthermore, the functionality of the components of the electronic device 600 can be implemented as a single computational cluster or multiple computing machines, which can communicate via a communication connection. Thus, the electronic device 600 can operate in a networked environment via logical connections with one or more other servers, network personal computers (PCs), or other network nodes.
[0116] The input device 650 may be one or more input devices, such as a mouse, a keyboard, a tracking ball, etc. The output device 660 may be one or more output devices, such as a display, a speaker, a printer, etc. The electronic device 600 may communicate, as necessary, via the communication unit 640 with one or more external devices (not shown), such as a storage device, a display device, one or more devices for user interaction with the electronic device 600, or any device (e.g., a network card, a modem, etc.) for communication between the electronic device 600 and one or more other electronic devices. Such communication may be performed via an input / output (I / O) interface (not shown).
[0117] According to an exemplary embodiment of the present disclosure, a computer-readable storage medium is provided having computer-executable instructions stored thereon, the computer-executable instructions being executed by a processor to perform the above-mentioned method. According to an exemplary embodiment of the present disclosure, a computer program product is further provided, the computer program product being tangibly stored on a non-transitory computer-readable medium and including computer-executable instructions, the computer-executable instructions being executed by a processor to perform the above-mentioned method.
[0118] Aspects of the present disclosure will now be described with reference to flowcharts and / or block diagrams of methods, apparatus, devices and computer program products implemented in accordance with the present disclosure, where each block of the flowcharts and / or block diagrams, and combinations of blocks in the flowcharts and / or block diagrams, may all be implemented by computer readable program instructions.
[0119] These computer readable program instructions may be provided to a processing unit of a general purpose computer, special purpose computer or other programmable data processing device to produce an apparatus, and when these instructions are executed by the processing unit of the computer or other programmable data processing device, the apparatus realizes the functions / operations specified in one or more blocks in the flowcharts and / or block diagrams. These computer readable program instructions may be stored on a computer readable storage medium, and these instructions cause the computer, programmable data processing device and / or other apparatus to operate in a particular manner, and the computer readable medium on which the instructions are stored includes an article of manufacture having various instructions for implementing the functions / operations specified in one or more blocks in the flowcharts and / or block diagrams.
[0120] The computer readable program instructions are loaded into a computer, other programmable data processing apparatus, or other device and cause the computer, other programmable data processing apparatus, or other device to execute a series of operational steps to produce a computer implemented process, and the instructions executing on the computer, other programmable data processing apparatus, or other device implement the functions / acts specified in one or more blocks in the flowcharts and / or block diagrams.
[0121] The flowcharts and block diagrams in the accompanying drawings illustrate possible architectures, functions, and operations of systems, methods, and computer program products that may be implemented according to aspects of the present disclosure. In this regard, each block in the flowcharts or block diagrams illustrates a module, program segment, or part of an instruction, which includes one or more executable instructions for implementing a specified logical function. In some alternative embodiments, the functions attached to the blocks may occur in a different order than that illustrated in the accompanying drawings. For example, two consecutive blocks may be executed substantially in parallel, and depending on such functions, may be executed in a reverse order. Also, each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts may be realized by a system based on dedicated hardware that executes the specified functions or operations, or may be realized by a combination of dedicated hardware and computer instructions.
[0122] Although each aspect of the present disclosure has been described above, the above description is illustrative, not exhaustive, and is not limited to each of the disclosed aspects. Many modifications and changes will be apparent to those skilled in the art without departing from the scope and spirit of each of the described aspects. The terms used in this specification are selected to best explain the principles, practical applications, or improvements in technology in the marketplace of each aspect, or to allow those skilled in the art to understand each embodiment disclosed in this specification.
Claims
1. determining a plurality of events to be processed from the one or more media contents, at least two of the events to be processed having an event relationship; determining an additional event set from the plurality of events to be processed based on an existing event set for constructing an event graph, and an additional event in the additional event set is different from an existing event in the existing event set; updating the event graph based on the additional event set.
2. Determining the additional event set includes: determining, for a given event in the plurality of events to be processed, a degree of correspondence between text describing the given event and text describing each existing event in the set of existing events; and The method of claim 1 , comprising identifying the given event as an additional event in response to all of the correspondence matches being less than a threshold match.
3. Identifying the given event as an additional event comprises: adding an item corresponding to the given event to a document-based database for the event graph to store information related to the given event; and 3. The method of claim 2, further comprising setting an additional event field in the item to a predetermined value indicating an additional event.
4. 3. The method of claim 2, further comprising cancelling the identification of the given event as an additional event in response to the update of the event graph.
5. 4. The method of claim 3, further comprising: in response to the degree of match between the text describing the given event and the text describing a first existing event exceeding the threshold degree of match, storing information related to a target media content among the one or more media contents in an item corresponding to the first existing event in the document-based database, wherein the given event is determined from the target media content.
6. Updating the event graph comprises: for a first additional event in the set of additional events, determining a first text element in a text describing the first additional event; updating element occurrence frequency information stored in a key-value database and corresponding to the first text element; determining a second existing event in the set of existing events, wherein text describing the second existing event includes the first text element; and The method of claim 1 , further comprising: updating a correlation degree of an event relationship between the second existing event and a third existing event in the event graph based on the updated element occurrence frequency information.
7. Updating the correlation degree includes: determining a first event having an event relationship with the first additional event in the set of existing events or the set of additional events, and a second text element in a text describing the first event; updating element-pair frequency-of-occurrence information stored in the key-value database that indicates that the first text element occurs in relation to the second text element; and 7. The method of claim 6, further comprising, in response to the text describing the third existing event including the second text element, updating the correlation degree based on updated element-pair occurrence frequency information.
8. for a second additional event in the set of additional events, determining a similarity between the second additional event and a second event in the set of additional events or the set of existing events; 2. The method of claim 1, further comprising: in response to the similarity exceeding a threshold similarity, storing an indication of a similarity relationship between the second additional event and the second event in a document-based database for the matter graph.
9. Updating the event graph comprises: determining an abstract event for summarizing the second additional event and at least the second event having a similarity relationship with the second additional event; mapping the second additional event and the second event to the abstract event in the event graph; and 9. The method of claim 8, further comprising adding nodes representing the abstract events to a visual representation of the event graph.
10. An electronic device comprising at least one processing circuit, said at least one processing circuit being configured to perform the method according to any one of claims 1 to 9.
Citation Information
Patent Citations
Log analysis system, log analysis method and log analysis device
JP2018022248A
Graph generation method, information processing apparatus and graph generation program
JP2019117442A
Ft diagram update device, flow diagram update device for trouble shooting, and program
JP2020160868A
Relevance analyzing device and method
JP2021103406A
Techniques for building knowledge graphs within limited knowledge domains
JP2021534493A
