CN116975290A - Data processing method, device, electronic equipment and computer readable storage medium - Google Patents
Data processing method, device, electronic equipment and computer readable storage medium Download PDFInfo
- Publication number
- CN116975290A CN116975290A CN202310900545.7A CN202310900545A CN116975290A CN 116975290 A CN116975290 A CN 116975290A CN 202310900545 A CN202310900545 A CN 202310900545A CN 116975290 A CN116975290 A CN 116975290A
- Authority
- CN
- China
- Prior art keywords
- transaction
- time period
- target
- features
- emotion
- 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.)
- Pending
Links
- 238000003672 processing method Methods 0.000 title claims abstract description 29
- 230000008451 emotion Effects 0.000 claims abstract description 279
- 230000002452 interceptive effect Effects 0.000 claims abstract description 79
- 238000012545 processing Methods 0.000 claims abstract description 48
- 238000000034 method Methods 0.000 claims description 59
- 230000003993 interaction Effects 0.000 claims description 30
- 238000012546 transfer Methods 0.000 claims description 26
- 238000012216 screening Methods 0.000 claims description 20
- 230000007704 transition Effects 0.000 claims description 18
- 230000004927 fusion Effects 0.000 claims description 13
- 238000006243 chemical reaction Methods 0.000 claims description 9
- 238000004590 computer program Methods 0.000 claims description 9
- 230000001902 propagating effect Effects 0.000 claims description 4
- 239000000284 extract Substances 0.000 description 19
- 230000001364 causal effect Effects 0.000 description 16
- 230000006870 function Effects 0.000 description 16
- 239000010410 layer Substances 0.000 description 16
- 238000005516 engineering process Methods 0.000 description 14
- 238000013473 artificial intelligence Methods 0.000 description 11
- 230000000694 effects Effects 0.000 description 11
- 239000011159 matrix material Substances 0.000 description 10
- 230000007935 neutral effect Effects 0.000 description 9
- 238000012549 training Methods 0.000 description 8
- 230000004913 activation Effects 0.000 description 7
- 238000013527 convolutional neural network Methods 0.000 description 7
- 238000000605 extraction Methods 0.000 description 7
- 238000003058 natural language processing Methods 0.000 description 7
- 230000008569 process Effects 0.000 description 7
- 230000000630 rising effect Effects 0.000 description 7
- 238000013528 artificial neural network Methods 0.000 description 6
- 238000007726 management method Methods 0.000 description 6
- 238000011161 development Methods 0.000 description 5
- 238000010586 diagram Methods 0.000 description 5
- 238000004891 communication Methods 0.000 description 4
- 230000002996 emotional effect Effects 0.000 description 4
- 239000002356 single layer Substances 0.000 description 4
- 238000004458 analytical method Methods 0.000 description 3
- 238000011156 evaluation Methods 0.000 description 3
- 230000006872 improvement Effects 0.000 description 3
- 230000009467 reduction Effects 0.000 description 3
- 238000013145 classification model Methods 0.000 description 2
- 230000007423 decrease Effects 0.000 description 2
- 230000007246 mechanism Effects 0.000 description 2
- 230000003287 optical effect Effects 0.000 description 2
- 230000000644 propagated effect Effects 0.000 description 2
- 230000002787 reinforcement Effects 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 238000012360 testing method Methods 0.000 description 2
- 230000001133 acceleration Effects 0.000 description 1
- 230000009471 action Effects 0.000 description 1
- 238000012098 association analyses Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000033228 biological regulation Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 238000004422 calculation algorithm Methods 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 230000015556 catabolic process Effects 0.000 description 1
- 238000004140 cleaning Methods 0.000 description 1
- 230000010485 coping Effects 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000006731 degradation reaction Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000007599 discharging Methods 0.000 description 1
- 235000013305 food Nutrition 0.000 description 1
- 238000002372 labelling Methods 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000012423 maintenance Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
- 239000010970 precious metal Substances 0.000 description 1
- 238000013519 translation Methods 0.000 description 1
- 235000013311 vegetables Nutrition 0.000 description 1
Classifications
-
- 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
- G06F16/35—Clustering; Classification
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/042—Knowledge-based neural networks; Logical representations of neural networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N5/00—Computing arrangements using knowledge-based models
- G06N5/04—Inference or reasoning models
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Computational Linguistics (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- Computing Systems (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- Databases & Information Systems (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The embodiment of the invention discloses a data processing method, a data processing device, electronic equipment and a computer readable storage medium; after digital transaction data, content texts and interactive texts of at least one transaction object in a plurality of time periods are obtained, transaction events are extracted from the content texts, text reasoning is carried out on the transaction events by adopting a preset rational map to obtain content characteristics of the transaction objects, digital characteristics are extracted from the digital transaction data, the content characteristics and the digital characteristics are fused to obtain object characteristics of each transaction object, emotion characteristics are extracted from the interactive texts, target relation characteristics among the transaction objects are determined according to characteristic relations among the emotion characteristics of different transaction objects, the target relation characteristics and the object characteristics are spliced to obtain target object characteristics of each transaction object, and transaction trend of each transaction object is predicted based on the target object characteristics; the scheme can improve the accuracy of data processing.
Description
Technical Field
The present application relates to the field of data processing, and in particular, to a data processing method, apparatus, electronic device, and computer readable storage medium.
Background
In recent years, with the rapid development of internet technology, a large amount of data is generated on a network. The massive data often contain a large amount of information, and the data can be processed in order to improve the utilization rate of the data. By taking the example of transaction scenario, the object data of the transaction object is processed, so that the future transaction trend of the transaction object can be predicted. In a transaction scene, the current data processing mode often collects data of two modes, namely transaction data and text data of a transaction object, carries out momentum overflow modeling, and predicts transaction trend based on the model.
In the course of research and practice of the current technology, the inventor of the present application found that the text data used by momentum overflow modeling are usually neutral content texts such as news texts describing transaction objects, and the emotional tendency of these neutral content texts is often not obvious, so that a certain error exists in the predicted transaction tendency, thus resulting in lower accuracy of data processing.
Disclosure of Invention
The embodiment of the invention provides a data processing method, a data processing device, electronic equipment and a computer readable storage medium, which can improve the accuracy of data processing.
A data processing method, comprising:
acquiring object data of at least one transaction object in a plurality of time periods, wherein the object data comprises digital transaction data and a plurality of text data, and the text data comprises content text and interactive text;
extracting transaction events from the content text, and carrying out text reasoning on the transaction events by adopting a preset rational map so as to obtain the content characteristics of each transaction object;
extracting digital characteristics from the digital transaction data, and fusing the content characteristics and the digital characteristics of the transaction objects to obtain object characteristics of each transaction object;
extracting emotion characteristics from the interactive text, and determining target relationship characteristics among the transaction objects according to characteristic relationships among the emotion characteristics of different transaction objects;
and splicing the target relation characteristic with the object characteristic to obtain a target object characteristic of each transaction object, and predicting the transaction trend of each transaction object based on the target object characteristic.
Accordingly, an embodiment of the present invention provides a data processing apparatus, including:
an acquisition unit, configured to acquire object data of at least one transaction object in a plurality of time periods, where the object data includes digital transaction data and a plurality of text data, and the text data includes content text and interactive text;
the reasoning unit is used for extracting transaction events from the content text and carrying out text reasoning on the transaction events by adopting a preset event map so as to obtain the content characteristics of each transaction object;
the fusion unit is used for extracting digital characteristics from the digital transaction data, and fusing the content characteristics and the digital characteristics of the transaction objects to obtain object characteristics of each transaction object;
the determining unit is used for extracting emotion characteristics from the interactive text and determining target relationship characteristics among the transaction objects according to characteristic relationships among the emotion characteristics of different transaction objects;
and the prediction unit is used for splicing the target relation characteristic with the object characteristic to obtain a target object characteristic of each transaction object, and predicting the transaction trend of each transaction object based on the target object characteristic.
In some embodiments, the inference unit may be specifically configured to propagate the transaction event in a preset event map, so as to obtain an event propagation value of each content text, where the event propagation value characterizes a logical relationship between the transaction event and a transaction trend; calculating the average value of the event propagation values of each time period in the event propagation values to obtain the initial content characteristics of the transaction object in each time period; and adjusting the initial content characteristics based on the time sequence information of the time period to obtain the content characteristics of each transaction object.
In some embodiments, the inference unit may be specifically configured to identify transition probabilities between the nodes in a preset event map; based on the event type of the transaction event, selecting a target event node corresponding to the transaction event from the event nodes; and according to the transition probability, the transaction event is transferred from the target event node in the preset event map so as to obtain an event propagation value of each content text.
In some embodiments, the inference unit may be specifically configured to transfer the transaction event from the target event node to the termination node in the preset event map according to the transfer probability; stopping transferring the transaction event when the transaction event reaches any one termination node, and taking the reached termination node as a target termination node; and determining an event propagation value of the content text based on the event type corresponding to the target termination node.
In some embodiments, the inference unit may be specifically configured to determine a target time period in the time periods, and screen out target initial content features of the target time period from the initial content features; adjusting the target initial content characteristics based on the time sequence information of the time period to obtain the current content characteristics of the target time period; returning to the step of determining the target time period in the time period until the time period is the target time period, and obtaining the current content characteristics of each time period; and taking the current content characteristic of each time period as the content characteristic of the transaction object.
In some embodiments, the inference unit may be specifically configured to screen at least one historical time period corresponding to the target time period from the time periods based on timing information of the time periods; extracting initial content characteristics corresponding to the historical time period from the initial content characteristics to obtain historical content characteristics; acquiring data weight corresponding to the target initial content characteristics, and determining adjustment weight corresponding to the historical content characteristics based on the data weight; and weighting the historical content characteristics according to the adjustment weight, and fusing the weighted historical content characteristics with the target initial content characteristics to obtain the current content characteristics of the target time period.
In some embodiments, the inference unit may be specifically configured to obtain a historical transaction data set of a preset historical time range; extracting target transaction data corresponding to a plurality of preset transaction events from the historical transaction data set, and determining a logic relationship between the preset transaction events based on the target transaction data; based on the logical relationship, constructing a rational map by taking the preset transaction event as a node, and taking the rational map as a preset rational map.
In some embodiments, the inference unit may be specifically configured to perform feature extraction on the content text to obtain text features of the content text; classifying the content text based on the text characteristics to obtain transaction event types corresponding to the content text; and screening out a preset transaction event corresponding to the transaction event type from the preset transaction event to obtain a transaction event corresponding to the content text.
In some embodiments, the fusing unit may be specifically configured to screen, based on a time period corresponding to the digital sub-feature, a target content feature corresponding to the digital sub-feature from the content features; splicing the digital sub-features and the target content features of each time period to obtain interaction features of each transaction object in a plurality of time periods; and extracting time sequence features from the interaction features, and taking the time sequence features as object features of the transaction objects.
In some embodiments, the determining unit may be specifically configured to perform emotion classification on the interactive text, and count the number of texts in different emotion categories in each time period; determining emotion tendency parameters corresponding to the transaction objects in each time period based on the text quantity, wherein the emotion tendency parameters indicate emotion tendency degrees of interaction aiming at the interaction objects; and carrying out feature coding on the emotion tendency parameters to obtain emotion features of the transaction object in each time period.
In some embodiments, the determining unit may be specifically configured to determine a positive emotion parameter of each time period based on the number of positive texts; according to the number of the negative texts, determining the negative emotion parameters of each time period; and calculating the ratio between the positive emotion parameters and the negative emotion parameters in the same time period to obtain emotion tendency parameters corresponding to the transaction objects in each time period.
In some embodiments, the determining unit may be specifically configured to identify at least one historical time period corresponding to each time period in the time periods, to obtain a target historical time period of each time period; screening emotion tendency parameters corresponding to the target historical time period from the emotion tendency parameters to obtain historical emotion tendency parameters of each time period; and carrying out feature coding on the emotion tendency parameters and the historical emotion tendency parameters of each time period to obtain emotion features of each time period.
In some embodiments, the determining unit may be specifically configured to screen emotion features of different transaction objects in each time period from the emotion features, to obtain an emotion feature set corresponding to each time period; extracting relation features among emotion features from the emotion feature set to obtain current relation features among the transaction objects in each time period; and adjusting the current relation characteristic based on the time sequence information of the time period to obtain a target relation characteristic corresponding to each time period.
In some embodiments, the determining unit may be specifically configured to fuse emotion features in the emotion feature set to obtain fused emotion features; extracting relation features from the fused emotion features, wherein the relation features represent correlations among the emotion features in the emotion feature set; and performing feature conversion on the relation features to obtain the current relation features among the transaction objects in each time period.
In some embodiments, the determining unit may be specifically configured to obtain a history relationship feature between the transaction objects in a preset history time period, and screen a candidate time period adjacent to the preset history time period from the time periods; fusing the current relation characteristic corresponding to the candidate time period with the history relation characteristic to obtain a target relation characteristic corresponding to the candidate time period; taking the candidate time period as the preset historical time period, and taking the target relation characteristic as the historical relation characteristic of the preset historical time period; and returning to the step of screening out the candidate time periods adjacent to the preset historical time period in the time periods until the time periods are all the candidate time periods, and obtaining the target relation characteristic corresponding to each time period.
In some embodiments, the prediction unit may be specifically configured to determine, based on the target relationship feature, an attention weight of each transaction object in multiple dimensions; weighting the object characteristics of each transaction object according to the attention weight to obtain candidate object characteristics of each pair of transaction objects in multiple dimensions; and splicing the candidate object features to obtain the target object features of each transaction object.
In addition, the embodiment of the application also provides electronic equipment, which comprises a processor and a memory, wherein the memory stores application programs, and the processor is used for running the application programs in the memory to realize the data processing method provided by the embodiment of the application.
In addition, the embodiment of the application also provides a computer readable storage medium, which stores a plurality of instructions, wherein the instructions are suitable for being loaded by a processor to execute the steps in any data processing method provided by the embodiment of the application.
In addition, the embodiment of the application also provides a computer program product, which comprises a computer program or instructions, and the computer program or instructions realize the steps in the data processing method provided by the embodiment of the application when being executed by a processor.
After object data of at least one transaction object in a plurality of time periods are acquired, the object data comprises digital transaction data and a plurality of text data, the text data comprises content text and interactive text, transaction events are extracted from the content text, text reasoning is carried out on the transaction events by adopting a preset event map to obtain content characteristics of each transaction object, then the digital characteristics are extracted from the digital transaction data, the content characteristics and the digital characteristics of each transaction object are fused to obtain object characteristics of each transaction object, emotion characteristics are extracted from the interactive text, target relation characteristics among the transaction objects are determined according to characteristic relations among the emotion characteristics of different transaction objects, then the target relation characteristics and the object characteristics are spliced to obtain target object characteristics of each transaction object, and transaction trend of each transaction object is predicted based on the target object characteristics; according to the scheme, aiming at neutral content texts, transaction events are extracted from the content texts, and text reasoning is carried out on the transaction events by adopting a preset rational map, so that causal relations in a transaction market are captured, more reliable logic information is provided for predicting transaction areas of transaction objects, interactive texts are introduced, emotion features are extracted from the interactive texts, and the current relation among the transaction objects is determined based on the emotion features, so that the transfer effect of emotion of the transaction objects or the transaction market can be captured, and further, the accuracy of predicting transaction trends is improved, and therefore, the accuracy of data processing can be improved.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings that are needed in the description of the embodiments will be briefly described below, it being obvious that the drawings in the following description are only some embodiments of the present invention, 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 schematic view of a scenario of a data processing method according to an embodiment of the present invention;
FIG. 2 is a flow chart of a data processing method according to an embodiment of the present invention;
FIG. 3 is a schematic diagram of a rational map provided by an embodiment of the present invention;
FIG. 4 is a schematic diagram of an overall framework for predicting stock exchange trends provided by an embodiment of the present invention;
FIG. 5 is another flow chart of a data processing method according to an embodiment of the present invention;
FIG. 6 is a schematic diagram of a data processing apparatus according to an embodiment of the present invention;
fig. 7 is a schematic structural diagram of an electronic device according to an embodiment of the present invention.
Detailed Description
The following description of the embodiments of the present invention will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present invention, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to fall within the scope of the invention.
The embodiment of the application provides a data processing method, a data processing device, electronic equipment and a computer readable storage medium. The data processing device may be integrated in an electronic device, which may be a server or a device such as a terminal.
The server may be an independent physical server, a server cluster or a distributed system formed by a plurality of physical servers, or a cloud server providing cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, network acceleration services (Content Delivery Network, CDN), basic cloud computing services such as big data and an artificial intelligent platform. The terminal may be, but is not limited to, a smart phone, a tablet computer, a notebook computer, a desktop computer, a smart speaker, a smart watch, etc. The terminal and the server may be directly or indirectly connected through wired or wireless communication, and the present application is not limited herein.
For example, referring to fig. 1, taking the case that the data processing apparatus is integrated in an electronic device, after obtaining object data of at least one transaction object in a plurality of time periods, the object data includes digital transaction data and a plurality of text data, the text data includes content text and interactive text, a transaction event is extracted from the content text, text reasoning is performed on the transaction event by adopting a preset event map to obtain content features of each transaction object, then the digital features are extracted from the digital transaction data, the content features and the digital features of the transaction object are fused to obtain object features of each transaction object, emotion features are extracted from the interactive text, and target relationship features between the transaction objects are determined according to feature relationships between the emotion features of different transaction objects, then the target relationship features and the object features are spliced to obtain target object features of each transaction object, and based on the target object features, transactions of each transaction object are predicted, so that data processing accuracy is improved.
The data processing method provided by the embodiment of the application relates to a natural language processing direction in artificial intelligence. According to the embodiment of the application, the object data of at least one transaction object in a plurality of time periods can be obtained, the object data can comprise digital transaction data and a plurality of text data, and the transaction trend of each transaction object is predicted by carrying out characteristics on the data transaction data and the text data and determining the target object characteristics of each transaction object, so that the accuracy rate of data processing is improved.
Among these, artificial intelligence (Artificial Intelligence, AI) is the theory, method, technique and application system that uses a digital computer or a machine controlled by a digital computer to simulate, extend and expand human intelligence, sense the environment, acquire knowledge and use knowledge to obtain optimal results. In other words, artificial intelligence is an integrated technology of computer science that attempts to understand the essence of intelligence and to produce a new intelligent machine that can react in a similar way to human intelligence. Artificial intelligence, i.e. research on design principles and implementation methods of various intelligent machines, enables the machines to have functions of sensing, reasoning and decision.
The artificial intelligence technology is a comprehensive subject, and relates to the technology with wide fields, namely the technology with a hardware level and the technology with a software level. Artificial intelligence infrastructure technologies generally include, for example, sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, big data processing technologies, pre-training model technologies, operation/interaction systems, mechatronics, and the like. The pre-training model is also called a large model and a basic model, and can be widely applied to all large-direction downstream tasks of artificial intelligence after fine adjustment. The artificial intelligence software technology mainly comprises a computer vision technology, a voice processing technology, a natural language processing technology, machine learning/deep learning and other directions.
Among them, natural language processing (Nature Language processing, NLP) is an important direction in the fields of computer science and artificial intelligence. It is studying various theories and methods that enable effective communication between a person and a computer in natural language. The natural language processing relates to natural language, namely the language used by people in daily life, and is closely researched with linguistics; while involving computer science and mathematics, and the like. An important technique for model training in the artificial intelligence domain, a pre-training model, is developed from a large language model (Large Language Model) in the NLP domain. Through fine tuning, the large language model can be widely applied to downstream tasks. Natural language processing techniques typically include text processing, semantic understanding, machine translation, robotic questions and answers, knowledge graph techniques, and the like.
It will be appreciated that in particular embodiments of the present application, where digital transaction data, text data, or other subject data is involved, where the following examples of the present application are applied to particular products or technologies, permissions or consents need to be obtained, and where the collection, use and processing of relevant data is required to comply with relevant national and regional laws and regulations and standards.
The following will describe in detail. The following description of the embodiments is not intended to limit the preferred embodiments.
The present embodiment will be described from the point of view of a data processing apparatus, which may be integrated in an electronic device, and the electronic device may be a server or a device such as a terminal; the terminal may include a tablet computer, a notebook computer, a personal computer (PC, personal Computer), a wearable device, a virtual reality device, or other devices capable of performing data processing.
A data processing method, comprising:
obtaining object data of at least one transaction object in a plurality of time periods, wherein the object data comprises digital transaction data and a plurality of text data, the text data comprises content text and interactive text, transaction events are extracted from the content text, text reasoning is carried out on the transaction events by adopting a preset event map so as to obtain content characteristics of each transaction object, digital characteristics are extracted from the digital transaction data, the content characteristics and the digital characteristics of each transaction object are fused to obtain object characteristics of each transaction object, emotion characteristics are extracted from the interactive text, target relation characteristics among the transaction objects are determined according to characteristic relations among the emotion characteristics of different transaction objects, the target relation characteristics are spliced with the object characteristics to obtain target object characteristics of each transaction object, and transaction trend of each transaction object is predicted based on the target object characteristics.
As shown in fig. 2, the specific flow of the data processing method is as follows:
101. object data of at least one transaction object over a plurality of time periods is acquired.
The transaction object may be an object capable of performing a transaction in a transaction market, and the transaction object may be an object of a entity or a virtual object. The types of trading objects may be varied and may include, for example, stocks, funds, bonds, notes, foreign exchange, precious metals, futures, spot or other objects that may be traded in the trading market, etc.
Wherein the object data may include digital transaction data and a plurality of text data, and the text data may include content text and interactive text. By digital transaction data is also understood the index data of the transaction object, which is often presented in digital or other statistically acceptable form, and the type of digital transaction data may be varied, such as transaction price, transaction amount, transaction period, transaction time, transaction cost, transaction profit, transaction code or other digital data indicative of the transaction index. Taking a stock as an example, the digital trading data can be a stock price (a price of a running, a price of a receiving, a highest price or a lowest price, etc.), a volume of a transaction, or index data of other stocks, etc. The content text may be text describing the transaction object, and is typically neutral text. The type of content text may be varied, and may include news, discipline, or other text that neutral describes the transaction object, for example. The interactive text may be text with emotion tendencies for the transaction objects, and the interactive text may be of various types, for example, may include comment text, comment text or other text with emotion tendencies, and the like.
The time period may be a period in which transaction data and text data are counted or collected. Each time period has the same time distance or length. The object data for different time periods may be the same or different. The type of time period may be varied and may include, for example, years, 1 year, half year, 3 months, 1 month, half month, one week, 1 day, 12 hours, 6 hours, 3 hours, 1 hour, 45 minutes, 30 minutes, 15 minutes, 10 minutes, 5 minutes, 1 minute, 30 seconds, or any other length of time. The time period may be set according to the actual application, may be set according to the transaction period, the transaction time, the business hours of the transaction market, or the like of the transaction object, or may be set according to the object identifier, the object type, or other attribute information of the transaction object. Taking a time period of 1 day as an example, the object data may include digital transaction data and text data of the transaction object within 1 day. It should be noted that, the end time of the time period may be the current time or a historical time before the current time, and the instant time period may include the current time period or the historical time period.
The method for acquiring the object data of at least one transaction object in a plurality of time periods may be various, and specifically may be as follows:
for example, the object data of at least one transaction object sent by the terminal in a plurality of time periods may be directly received, or the current transaction data and the historical transaction data of the at least one transaction object may be extracted in the transaction platform or the transaction database, the digital transaction data of the plurality of time periods may be extracted in the current transaction data and the historical transaction data, respectively, the plurality of content texts and the plurality of interactive texts of the at least one transaction object in the plurality of time periods may be extracted in the network or the text database, thereby obtaining text data of the transaction object, the text data and the digital transaction data may be used as the object data of the transaction object in the plurality of time periods, or when the number of the object data is large or the occupied memory is large, a data processing request may be received, the data processing request carries a storage address of the object data of the at least one transaction object in the plurality of time periods, the object data of the plurality of time periods may be obtained based on the storage address, and so on.
102. And extracting transaction events from the content text, and carrying out text reasoning on the transaction events by adopting a preset rational map so as to obtain the content characteristics of each transaction object.
The transaction event may be an event representing content text represented in the transaction or financial field, and may also be referred to as a financial event. The types of the transaction event (financial event) may be various, for example, taking a transaction object as a stock, and may include a Equity Transfer (Equity Transfer), an Investment (Investment), a control share (suse), a reduction hold (Reduce hold), a Guarantee (guard), a contract, a market judgment (juddement), an Increase hold (Increase hold), a stock Increase (Stock price Increase), a stock drop (Stock price decreases), a Default Type (Default Type), a product management problem (Product management issues), a company bulletin (Company announcement), a product performance improvement (Performance Increase), a product performance drop (Performance decrease), a Policy support (Policy support) or a Policy opposition (Policy support), and the like. By default type, it is understood that transaction events that do not belong to any class are categorized into this class. The transaction events corresponding to different transaction objects may be the same or different, i.e., the transaction events corresponding to other transaction objects may include any one or more of the 17 transaction events, may also include other transaction events, and so on.
The preset rational atlas may be a preset or constructed rational atlas for transaction events. The so-called event map may be a logical knowledge base of events describing the rules and patterns of evolution between events. Structurally, the event map is a directed graph, and the directed graph comprises nodes and directed edges, wherein the nodes represent events, and the directed edges represent the following, causal, conditional, upper and lower event logic relations among the events. The cis-relationship means that the occurrence of the former event (cause) causes the occurrence of the latter event (result) between two events. The causal relationship satisfies the partial order relationship of the cause event before and the result event after in time, so that the causal relationship can be considered as a subset of the compliant relationship in a certain sense. There is a causal intensity value between the causal event pairs between 0 and 1, indicating the confidence with which the causal relationship holds. The causal relationship means that the former event is a condition under which the latter event occurs. The conditional relationships belong to a certain logical relationship of propositions in ideas, and the causal relationships belong to a certain knowledge of objective facts. The conditional relationships can be understood as "reasons" as the inherent links of premises to conclusions or arguments to arguments, as to logic, while the causal relationships can be understood as "reasons" as to facts, essentially "reasons not equal. For example, the condition of "if there are many persons buying a ticket, the movie is good" is true, and the cause of "because there are many persons buying a ticket, the movie is good" is not true. The term upper and lower relationships may include term upper and lower relationships and action upper and lower relationships, for example, the event "food price increases" and "vegetable price increases" are mutually term upper and lower relationships; the events 'fusion' and 'splicing' are mutually in an upper-lower relationship of the part of speech. It should be noted that the context is generally deterministic, so there is no need to analogize to compliance or cause and effect relationships, and a constant between 0 and 1 is assigned to the context to indicate its confidence.
The transaction event is extracted from the content text, and a preset rational map is adopted to perform text reasoning on the transaction event, so as to obtain the content characteristics of each transaction object, which can be specifically as follows:
s1, extracting a transaction event from the content text.
For example, feature extraction may be performed on the content text to obtain text features of the content text, and the content text is classified based on the text features to obtain transaction event types corresponding to the content text, and preset transaction events corresponding to the transaction event types are screened out from the preset transaction events to obtain transaction events corresponding to the content text.
Wherein the text feature may be feature information indicating a content type of the content text. The feature extraction method for the content text may be various, for example, the feature encoding may be performed on the content text by using a trimmed BERT (a bi-directional encoding network) to obtain text features of the content text, or the feature encoding may be performed on the content text by using other encoding networks capable of performing content encoding to obtain text features of the content text, or the like.
After extracting the characteristics of the content text, the content text can be classified based on the extracted text characteristics, so that the transaction event type corresponding to the content text is obtained. The transaction event type may be a type indicating an event to which the content text corresponds in the transaction or financial domain. Each transaction event type may correspond to a preset transaction event. The preset transaction event may be a transaction event corresponding to each node in the preset event map. The text feature may be used to classify the content text in various manners, for example, the text feature may be matched with an event feature corresponding to each transaction event type, the transaction event type corresponding to the event feature that is successfully matched is used as the transaction event type corresponding to the content text, or the event feature corresponding to each transaction event may be obtained, the feature similarity or the feature distance between the text feature and the event feature may be calculated, and the transaction event type corresponding to the content text may be determined based on the feature similarity or the feature distance, or the content text may be classified by using a classification network based on the text feature, to obtain the transaction event type corresponding to the content text, and so on.
After classifying the content text, the preset transaction event corresponding to the transaction event type can be screened out from the preset transaction events, so that the transaction event corresponding to the content text is obtained.
S2, carrying out text reasoning on the transaction event by adopting a preset event map so as to obtain the content characteristics of each transaction object.
For example, the transaction event may be propagated in a preset event map to obtain an event propagation value of each content text, and an average value of the event propagation values of each time period is counted in the event propagation statistics value to obtain an initial content feature of the transaction object in each time period, and the initial content feature is adjusted based on time sequence information of the time period to obtain a content feature of each transaction object, which may be specifically as follows:
(1) And transmitting the transaction event in a preset event map to obtain an event transmission value of each content text.
The preset event map comprises a plurality of nodes with logic relations, and the nodes comprise event nodes. The event propagation value characterizes a logical relationship between the transaction event and the transaction trend. Different transaction trends may correspond to different event propagation values.
The manner of propagating the transaction event in the preset event map may be various, and specifically may be as follows:
for example, the transition probability between nodes can be identified in a preset event map, a target event node corresponding to the transaction event is screened out from the event nodes based on the event type of the transaction event, and the transaction event is transferred from the target event node in the preset event map according to the transition probability, so as to obtain the event propagation value of each content text.
The transition probability may be understood as the probability of a transaction event being transitioned from one node to another. The method for identifying the transition probability between the nodes in the preset event map may be various, for example, the weight corresponding to the node edge between the nodes may be identified in the preset event map, and the weight is used as the probability of the directed transition between the two corresponding nodes.
After the transition probabilities among the nodes are identified and the target event nodes corresponding to the transaction events are screened, the transaction events can be transferred from the target event nodes in a preset event map according to the transition probabilities, so that the event propagation value of each content text is obtained. The nodes in the preset rational map may also comprise a plurality of end nodes, the so-called end nodes comprising nodes indicating a tendency for the transaction. The termination node is a node for terminating reasoning in the preset theory map, namely, after the transaction event in the preset theory map is transferred to the termination node, the transfer can be stopped. The manner of transferring the transaction event from the target event node in the preset event map may be various according to the transfer probability, for example, the transaction event may be transferred from the target event node to the termination node in the preset event map according to the transfer probability, when the transaction event reaches any one termination node, the transfer of the transaction event is stopped, the reached termination node is used as the target termination node, and the event propagation value of the content text is determined based on the event type corresponding to the target termination node.
The manner of transferring the transaction event from the target event node to the termination node in the preset event map may be various according to the transfer probability, for example, the transaction event may be randomly walked from the target event node to the termination node in the preset event map according to the transfer probability, or at least one transfer path of the target event node to the termination node may be determined in the preset event map according to the transfer probability, and based on the transfer path, the transaction event may be transferred or walked from the target event node in the preset event map.
When the transaction event reaches any one of the termination nodes, the transfer of the transaction event can be stopped, and the reached termination node is taken as a target termination node. Here, any one of the terminating nodes that arrives means that the terminating node is the terminating node that the transaction event arrives first or arrives at the shortest time.
After stopping the transfer of the transaction event and taking the reached termination node as the target termination node, the event propagation value of the content text can be determined based on the event type corresponding to the target termination node. The method for determining the event propagation value of the content text may be various, for example, the event type corresponding to the target termination node is obtained, the preset event propagation value corresponding to the event type is screened out from the preset event propagation value set, for example, taking the trading object as a stock as an example, the event type corresponding to the termination node may include two types of stock rising and stock falling, when the event type corresponding to the target termination node is stock rising, the preset event propagation value (for example, may be 1 or any other value) corresponding to stock rising may be screened out from the preset event propagation value set as the event propagation value of the content text, when the event type corresponding to the target termination node is stock falling, the preset event propagation value corresponding to falling may be screened out from the preset event propagation value set (for example, may be-1 or any other value but different from the preset event propagation value corresponding to stock rising) as the event propagation value of the content text, and so on.
In some embodiments, the transaction event is propagated in a preset event map, and the preset event map may be further constructed before the event propagation value of each content text is obtained. There are various ways of constructing the preset event map, for example, a historical transaction data set of a preset historical time range may be obtained, a plurality of target transaction data corresponding to the preset transaction time are extracted from the historical transaction data set, a logic relationship between preset events is determined based on the target transaction data, the transaction events are used as nodes to construct the event map based on the logic relationship, and the event map is used as the preset event map.
The historical transaction data set may include historical data including various types of transaction or financial fields, among others. Logical relationships may include compliance, cause and effect, condition, and upper and lower, etc., and may be described in detail above. The constructed rational atlas essentially describes the development rules of various transaction events (financial events). Taking a trade object as a stock, a preset trade event is a common trade event of 17 types, such as share right transfer, investment, control, deduction, guarantee, contract, judgment, maintenance, stock rising, stock falling, default type, product management problem, company notice, product performance improvement, product performance reduction, policy support or policy opposition, etc., as an example, the constructed event map can be as shown in fig. 3, and the node identification and the corresponding trade event can be as shown in table 1, and specifically as follows:
TABLE 1
| Node identification | Transaction event | Node identification | Transaction event |
| 0 | Equity transfer | 9 | Stock drop |
| 1 | Investment in | 10 | Default type |
| 2 | Strand control | 11 | Product management issues |
| 3 | Reduction of hold | 12 | Company bulletin |
| 4 | Guarantee for guarantee | 13 | Product performance improvement |
| 5 | Contract making | 14 | Product performance degradation |
| 6 | Market judgement | 15 | Policy support |
| 7 | Increase the holding | 16 | Policy opposition |
| 8 | Stock rising |
Wherein, the nodes 8 and 9 are termination nodes in the event map, and the other nodes are event nodes. The numbers of directed edges in the rational atlas represent weights corresponding to logical relationships transferred from one node to another according to the transfer direction.
(2) And calculating the average value of the time propagation values of each time period in the time propagation values to obtain the initial content characteristics of the transaction object in each time period.
For example, the time propagation value of each time period may be screened out from the time propagation values, a set of time propagation values corresponding to each time period is obtained, and a mean value of the time propagation values in the set of time propagation values is calculated, so as to obtain an initial content feature of each time period.
Taking the time period of 1 day as an example, the time propagation values of the content text of the same day can be screened out from the time propagation values, so as to obtain a time propagation value set corresponding to each day, and calculating the average value of the time propagation values in the time propagation value set, so as to obtain the initial content characteristics corresponding to each day.
(3) Based on the time sequence information of the time period, the initial content characteristics are adjusted to obtain the content characteristics of each transaction object.
For example, a target time period may be determined in the time period, a target initial content feature of the target time period may be selected from the initial content features, the target initial content feature may be adjusted based on time sequence information of the time period to obtain a current content feature of the target time period, and the step of determining the target time period in the time period is performed again until the time periods are all the target time periods, so as to obtain the current content feature of each time period, and the current content feature of each time period is used as the content feature of the transaction object.
The method for determining the target time period in the time periods can be various, for example, one time period can be selected randomly or randomly from the time periods as the target time period, or the target time period can be selected from the time periods based on time sequence information of the time periods.
After the target time period is determined, the target initial content characteristics of the target time period can be screened out from the initial content characteristics, and the target initial content characteristics are adjusted based on time sequence information of the time period, so that the current content characteristics of the target time period are obtained. The time sequence information of the time period can be understood as ordering information after ordering according to the time distance between the start-stop time and the current time of the time period. For example, the timing information may include day 1, day 2, or day n, for example, for a time period of one day. The adjustment manner of the target initial content may be various based on the time sequence information of the time period, for example, at least one historical time period corresponding to the target time period may be screened out in the time period based on the time sequence information of the time period, the initial content feature corresponding to the historical time period is extracted from the initial content feature to obtain the historical content feature, the data weight corresponding to the target initial content feature is obtained, the adjustment weight corresponding to the historical content feature is determined based on the data weight, the historical content feature is weighted according to the adjustment weight, and the weighted historical content feature and the target initial content feature are fused to obtain the current content feature of the target time period.
The time-series information of the time period may be used to screen at least one historical time period corresponding to the target time period from the time periods, for example, taking a preset time range corresponding to the time periods including the first 30 days from the current time, the time period being 1 day, in the time-series information, the 1 st day is substantially the 1 st day of the first 30 days, that is, the time distance from the current time is 30 days, the 2 nd day is substantially the 2 nd day of the first 30 days, that is, the time distance from the current time is 29 days, and so on, the 30 th day is substantially the 30 th day of the first 30 days, the time distance from the current time is 1 day, and taking the target time period as the d day, the at least one historical time period corresponding to the d day may be the (d-1) th to the 1 st day, that is, the total (d-1) historical time period.
After at least one historical time period corresponding to the target time period is screened out from the time periods, initial content characteristics corresponding to the historical time periods can be extracted from the initial content characteristics, and the historical content characteristics are obtained. And then, acquiring the data weight corresponding to the target initial content feature, and determining the adjustment weight corresponding to the historical content feature based on the data weight.
Wherein the adjustment weight may indicate a degree of influence of the historical content feature on the target initial content feature. The method for determining the adjustment weight corresponding to the historical content feature may be various based on the data weight, for example, based on the data weight, determining the initial adjustment weight corresponding to the historical content feature, calculating the time distance between each historical time period and the target time period to obtain the target time distance corresponding to each historical time period, and fusing the target time distance and the initial adjustment weight to obtain the adjustment weight corresponding to the historical content feature.
After the adjustment weight corresponding to the historical content feature is determined, the historical content feature can be weighted according to the adjustment weight, and the weighted historical content feature and the target initial content feature are fused to obtain the current content feature of the target time period. The weighted historical content features and the target initial content features may be fused in various manners, for example, the weighted historical content features and the target initial content features may be accumulated to obtain fused content features, the adjustment weights and preset fusion weights corresponding to the target initial content features are fused to obtain fused weights, and the ratio between the fused content features and the fused weights is calculated to obtain the current content features of the target time period. Taking the target time period as the d day as an example, the specific formula may be shown in formula (1), and the specific formula may be as follows:
Wherein t is d As a current content feature on day d,for the target initial content feature on day d, alpha is the data weight corresponding to the target initial content feature, +.>And d is the identification of the number of days, namely the time period, which is the historical content characteristic corresponding to the d-th day.
After the target initial content characteristics are adjusted, the step of determining the target time period in the time period can be carried out again until the time period is the target time period, so that the current content characteristics of each time period are obtained, the current content characteristics of each time period are used as the content characteristics of the transaction objects, and the content characteristics of each transaction object are obtained.
103. And extracting digital characteristics from the digital transaction data, and fusing the content characteristics and the digital characteristics of the transaction objects to obtain the object characteristics of each transaction object.
Wherein the object characteristics may be characteristic information characterizing at least one dimension of the transaction object.
Wherein the digital signature may be an index signature indicative of fluctuations or trends in the transaction process by the transaction object, and the digital signature may include digital sub-signatures for each time period.
The digital feature may be extracted from the digital transaction data in various ways, and specifically may be as follows:
For example, the digital transaction data may be feature-coded to obtain digital sub-features of each time period, and the digital sub-features are used as digital features, or digital data corresponding to the target transaction index may be extracted from the digital transaction data, and feature-coded or converted to obtain digital sub-features of each time period, and the digital sub-features are used as digital features, and so on.
The target trading index may be a trading index indicating or labeling a fluctuation or trend of a trading object in a trading process. Taking a stock as an example, the target trading index may be an index related to a stock price, for example, the index may include an index calculated by using the initial data related to the price of the stock, the highest price, the lowest price, the average stock price on the same day, the holding price, the clearing price or other prices, for example, SMA (simple moving average line), EMA (index moving average), MACD (heterogeneous moving average line), and the like.
After the digital features are extracted, the content features and the digital features of the transaction objects can be fused to obtain the object features of each transaction object. The method of fusing the content features and the digital features of the transaction objects may be various, for example, the method may be based on a time period corresponding to the digital sub-features, select a target content feature corresponding to the digital sub-features from the content features, splice the digital sub-features and the target content features of each time period to obtain an interaction feature of each transaction object in a plurality of time periods, extract a time sequence feature from the interaction feature, and use the time sequence feature as an object feature of the transaction object.
The manner of stitching the digital sub-feature of each time period with the target content feature may be various, for example, stitching the digital sub-feature of each time period with the target content feature to obtain a stitched object feature of each time period, multiplying the digital sub-feature of each time period with the target content feature to obtain a candidate object feature of each time period, and fusing the candidate object feature with the stitched object feature to obtain an interaction feature corresponding to each time period, so as to obtain interaction features of each object in multiple time periods, which may be shown in formula (2), specifically may be as follows:
wherein k is i For the interaction characteristics of the ith transaction object in a plurality of time periods, n i ∈R n For the digital character of the ith transaction object, t i ∈R t For the content characteristics of the ith transaction object, J ε R n×t×k 、W e ∈R k×(n+t) 、b∈R k Are all learnable parameters.
After the digital sub-features of each time period are spliced with the target content features to obtain the interactive features of each transaction object in a plurality of time periods, the time sequence features can be extracted from the interactive features. The manner of extracting the timing sequence feature from the interaction feature may be various, for example, a GRU (a gating loop network) may be used to extract the timing sequence feature from the interaction feature, which may be shown in the formula (3), and may be specifically as follows:
s i (L)＝GRU(K i (L)) (3)
Wherein s is i (L) is the time sequence characteristics of the ith transaction object in a plurality of time periods, K i (L) is an interactive feature for a plurality of time periods, for example, 1 day, K i (L)＝[k i (1)，k i (2)，…，k i (L)]，k i (1) For the interaction characteristics of the ith transaction object on day 1, k i (2) For the interaction characteristics of the ith transaction object on the 2 nd day, and so on, k i (L) is the interactive feature of the ith transaction object on the L th day.
In some embodiments, other timing feature extraction networks may also be employed to extract timing features from the interaction features.
After the time sequence feature is extracted from the interaction feature, the time sequence feature can be used as the object feature of the transaction object.
104. And extracting emotion characteristics from the interactive text, and determining target relationship characteristics among the transaction objects according to characteristic relationships among the emotion characteristics of different transaction objects.
Wherein, the emotion characteristics may be characteristic information indicating emotion tendencies for the transaction object in each time period.
The characteristic relation can be the interrelation between the emotion characteristics of different transaction objects.
Wherein the target relationship characteristic may be characteristic information indicating a relationship between different transaction objects over a plurality of time periods.
The method comprises the steps of extracting emotion characteristics from an interactive text, and determining target relationship characteristics of a transaction object according to characteristic relationships among the emotion characteristics of the transaction object, wherein the target relationship characteristics can be specifically as follows:
and C1, extracting emotion characteristics from the interactive text.
For example, the interactive text may be subjected to emotion classification, the text number of different emotion categories in each time period is counted, emotion tendency parameters corresponding to the transaction object in each time period are determined based on the text number, and feature encoding is performed on the emotion tendency parameters to obtain emotion characteristics of the transaction object in each time period, which may be specifically as follows:
(1) And carrying out emotion classification on the interactive texts, and counting the number of texts of different emotion types in each time period.
Wherein emotion categories may include positive emotion and negative emotion.
For example, the BERT model may be used to perform emotion classification on the interactive text, so as to obtain an emotion type of the interactive text, or other classification models (networks) that may perform emotion classification may be used to perform emotion classification on the interactive text, so as to obtain an emotion type of the interactive text, and so on.
After emotion classification is performed on the interactive text, the number of texts of different emotion categories in each time period can be counted. The number of texts of different emotion types in each time period can be counted in various manners, for example, interactive texts of the same emotion type in each time period are screened out from the interactive texts, so as to obtain a positive interactive text set corresponding to positive emotion in each time period and a negative interactive text set corresponding to negative emotion, the number of interactive texts is counted in the positive interactive text set, so as to obtain the number of positive texts in each time period, and the number of interactive texts is counted in the negative interactive text set, so as to obtain the number of negative texts in each time period.
(2) And determining emotion tendency parameters corresponding to the transaction objects in each time period based on the text quantity.
The emotion tendency parameter characterizes emotion tendency degree of the interactive object in a time period.
The manner of determining the emotion tendency parameter corresponding to the transaction object in each time period based on the text quantity may be various, and may specifically be as follows:
for example, the positive emotion parameters of each time period can be determined based on the number of positive texts, the negative emotion parameters of each time period are determined according to the number of negative texts, and the ratio between the positive emotion parameters and the negative emotion parameters of the same time period is calculated to obtain the emotion tendency parameters corresponding to the transaction objects in each time period. Taking a time period of 1 day as an example, the manner of calculating the emotion tendencies parameters corresponding to the transaction object may be as shown in the formula (4), and may specifically be as follows:
wherein e d For the emotional tendency parameters corresponding to the transaction object on the d day, pos is the positive text quantity on the d day, and neg is the negative text quantity on the d day.
(3) And carrying out feature coding on the emotion tendency parameters to obtain emotion features of the transaction object in each time period.
For example, at least one historical time period corresponding to each time period can be identified in the time periods to obtain a target historical time period of each time period, emotion trend parameters corresponding to the target historical time period are screened out from emotion trend parameters to obtain historical emotion trend parameters of each time period, and feature coding is performed on the emotion trend parameters of each time period and the historical emotion trend parameters to obtain emotion features of each time period.
The manner of identifying at least one historical time period corresponding to each time period in the time periods may be referred to above, and will not be described in detail herein.
After the target historical period corresponding to each time period is identified, the emotion tendency parameters corresponding to the target historical period can be screened out in the emotion tendency parameters so as to obtain the historical emotion tendency parameters of each time period.
After the historical emotion tendency parameters of each time period are screened out, feature encoding can be carried out on the emotion tendency parameters and the historical emotion tendency parameters of each time period, so that emotion features of each time period are obtained. The feature encoding manner of the emotion trend parameter and the historical emotion trend parameter of each time period may be various, for example, the GRU network may be used to encode the emotion trend parameter and the historical emotion trend parameter of each time period into a unified vector, the unified vector is taken as the emotion feature under the period time, the time period is 1 day, the time period is the d day, the plurality of time periods are the L days, and the emotion feature on the d day may be calculated as shown in the formula (5), and specifically may be as follows:
e i ＝GRU(e d ,e d-1 ,…,e d-L+1 ) (5)
Wherein e i For the emotional characteristics of the ith transaction object on the d day, e d For the emotional tendency parameter of the ith subject on day d, (e) d-1 …e d-L+1 ) Is the historical emotion tendency parameter corresponding to the d day.
And C2, determining target relation characteristics among the transaction objects according to characteristic relations among the emotion characteristics of different transaction objects.
For example, emotion characteristics of different transaction objects in each time period can be screened out from emotion characteristics to obtain emotion characteristic sets corresponding to each time period, relationship characteristics among the emotion characteristics are extracted from the emotion characteristic sets to obtain current relationship characteristics among the transaction objects in each time period, and the current relationship characteristics are adjusted based on time sequence information of the time period to obtain target relationship characteristics corresponding to each time period.
The method for extracting the relation features between the emotion features in the emotion feature set can be various, for example, the emotion features in the emotion feature set can be fused to obtain fused emotion features, the relation features are extracted from the fused emotion features, and the relation features are converted to obtain the current relation features of the transaction objects in each time period.
The method for fusing the emotion features in the emotion feature set can be various, for example, a broadcast operation mechanism can be adopted to broadcast and add the emotion features in the emotion feature set to obtain fused emotion features, or an attention network can be adopted to extract associated features of the emotion features in the emotion feature set, determine attention weights of the emotion features based on the associated features, weight the emotion features based on the attention weights, and fuse the weighted emotion features to obtain fused emotion features.
After the emotion features in the emotion feature set are fused, the relationship features can be extracted from the fused emotion features. So-called relational features may characterize the interrelationship between emotion features in a set of emotion features. The method for extracting the relational features from the fused emotion features can be various, for example, a one-dimensional convolutional neural network can be adopted to extract the relational features from the fused emotion features, or other feature extraction networks can be adopted to extract the relational features from the fused emotion features, and the like.
After the relation features are extracted from the fused emotion features, the relation features can be subjected to feature exchange, so that the current relation features among the transaction objects in each time period are obtained. The feature conversion method of the relation feature may be various, for example, at least one-dimensional convolutional neural network may be used to perform feature conversion on the relation feature to obtain a current relation feature between transaction objects in each time period, or other feature conversion networks may be used to perform feature conversion on the relation feature to obtain a current relation feature between transaction objects in each time period, and so on.
Taking the example of extracting the relation features between emotion features in an emotion feature set by adopting a two-layer one-dimensional convolutional neural network, the relation features can be extracted from the fused emotion features by adopting a first-layer one-dimensional convolutional neural network, and the relation features are subjected to feature conversion by adopting a second-layer one-dimensional convolutional neural network, so that the current relation features between transaction objects in each time period are obtained, an activation function corresponding to the two-layer one-dimensional convolutional neural network can be ELU (an activation function), and the relation features between the emotion features can be extracted from the emotion feature set as shown in a formula (6), and specifically can be as follows:
wherein out is the current relationship feature between transaction objects, b 1 And b 2 Bias (a convolution kernel parameter) representing the first layer and the second layer, respectively, W 1 And W is 2 Representing the convolution kernels of the first layer and the second layer, respectively, C1 and C2 being the number of input channels of the first layer and the second layer, respectively, and being convolution operations, e being the emotion feature.
After extracting the current relation features among the transaction objects in each time period, the current relation features can be adjusted based on the time sequence information of the time period to obtain the target relation features corresponding to each time period. The current relation feature may be adjusted in various ways, for example, a history relation feature between transaction objects in a preset history time period may be obtained, a candidate time period adjacent to the preset history time period is screened out in the time period, the current relation feature corresponding to the candidate time period and the history relation feature are fused to obtain a target relation feature corresponding to the candidate time period, the candidate time period is taken as the preset history time period, the target relation feature is taken as the history relation feature of the preset history time period, and the step of screening out the candidate time period adjacent to the preset history time period in the time period is performed until the time periods are all the candidate time periods, so as to obtain the target relation feature corresponding to each time period.
The preset historical time period may be a historical time period before a plurality of preset time periods, that is, the preset historical time period may be a historical time period before a plurality of time periods of the current data processing for the transaction object, or may be a time period corresponding to the last data processing for the transaction. The method of screening the candidate time periods of the candidate time periods adjacent to the preset time period may be various, for example, when the preset historical time period is not in a plurality of time periods, the time period with the longest time distance from the current time is screened in the time periods as the candidate time period adjacent to the preset historical time period, and when the preset historical time is in a plurality of time periods (i.e., the preset historical time period is one of a plurality of time periods), the time period adjacent to the preset historical time period is screened in the time periods, so as to obtain the candidate time period, for example, when the preset time period is the d day, the candidate time period may be the (d+1) th day.
After candidate time periods adjacent to the preset historical time period are screened out from the time periods, the current relation features corresponding to the candidate time periods and the historical relation features can be fused, so that target relation features corresponding to the candidate time periods are obtained. The manner of fusing the current relationship feature corresponding to the candidate time period with the history relationship feature may be various, for example, the current relationship feature corresponding to the candidate time period and the history relationship feature may be added to obtain a fused relationship feature, and the fused relationship feature is mapped to the target relationship feature through an activation function, and the time period is 1 day as an example, which may be specifically shown in the formula (7), and may be specifically as follows:
Wherein R is d As the target relation characteristic among the transaction objects on the d th day, the d th day is a candidate time period, out is the current relation characteristic corresponding to the candidate time period, R d-1 As the characteristic of the history relationship, the (d-1) th day can be a preset history time period, W s Is a parameter that can be learned.
After the current relation feature corresponding to the candidate time period is fused with the history relation feature, the candidate time period can be used as a preset history time period, the target relation feature is used as a history relation feature of the preset history time period, and then the step of screening out the candidate time period adjacent to the preset history time period in the time period is carried out again until the time periods are all candidate time periods, so that the target relation feature corresponding to each time period is obtained.
Considering that the relationship features (matrix) of the transaction objects in the past have a certain influence on the relationship features (matrix) of the present, therefore, in the scheme, the relationship features or the history relationship features generated last time are added with the current relationship features, and the final target relationship features (matrix) of each time period are generated through an activation function, so that a graph representing the relationship between the transaction objects can be dynamically constructed, the transfer effect of the emotion of the transaction market can be captured, and the transaction trend of the transaction objects can be predicted in different time periods, and therefore, the accuracy of data processing can be improved.
105. And splicing the target relation characteristic and the object characteristic to obtain the target object characteristic of each transaction object, and predicting the transaction trend of each transaction object based on the target object characteristic.
The trade trend may be a development trend of the trade object in a future time period (one or more future time periods of the current moment), for example, the trade object may include a stock up and a stock down, or may further include a stock up amplitude and a stock down amplitude, and so on.
The manner of splicing the target relationship feature and the object feature may be various, and specifically may be as follows:
for example, the attention weight of each transaction object in multiple dimensions may be determined based on the target relationship feature, the object feature of each transaction object may be weighted according to the attention weight to obtain candidate object features of each transaction object in multiple dimensions, the candidate object features may be spliced to obtain target object features of each transaction object, or a graph network between transaction objects in multiple time periods may be constructed based on the target relationship features, and the candidate object features in multiple dimensions may be extracted from the object features by using the graph network, and the candidate object features may be spliced to obtain target object features in each time period, or the object relationship features in multiple dimensions corresponding to each transaction object may be directly extracted from the target relationship features, and the target object features in each time period may be obtained by splicing the object relationship features and the object features.
The attention weight of each transaction object in a plurality of dimensions can be determined through the graph network, the object characteristics of each transaction object are weighted according to the attention weight, candidate object characteristics of each transaction object in the plurality of dimensions are obtained, and the candidate object characteristics are spliced to obtain target object characteristics of the transaction object in each time period.
The network structure of the graph network may be various, for example, may include GCN (a graph network), GAT (a graph ideographic network), or other graph networks. Taking a graph network as a GAT network as an example, m attention heads in the GAT network are adopted to generate m candidate object features based on target relation features and object features, and the m candidate object features are spliced, so that the target object features of a transaction object in each time period are obtained, and the time period is 1 day as an example, as shown in a formula (8), the following can be specifically adopted:
wherein z is i Target object feature for the ith transaction object, m is the number of attention heads, s i (L) is the object feature of the ith transaction object in L time periods, R d Is a target relationship feature on day d.
After the target relation features and the object features are spliced, the transaction trend of each transaction object can be predicted based on the spliced target object features. The manner of predicting the transaction trend of each transaction object based on the target object features may be various, for example, a single-layer prediction network may be used to predict the transaction trend of each transaction object based on the target object features, so as to obtain a predicted transaction trend of each object, which may be shown in formula (9), specifically may be as follows:
Wherein,,z is the transaction trend (i.e., the likelihood of future changes) of the ith transaction object i Target object feature, W, of the ith transaction object o And b is a learnable parameter.
Optionally, in some embodiments, other classification networks may be further used to predict the probability of each transaction object in a preset development trend based on the target object characteristics, so as to obtain the transaction trend of each transaction object.
Taking a time period of 1 day and a plurality of time periods of L days as an example, a trade object is a stock, a content text is a news text, an interactive text is a comment text, and digital trade data is trade index data, an overall framework for predicting a trade trend of the stock based on the news text, the comment text and the trade index data in the scheme may be as shown in fig. 4, and may include four parts of an index learning module based on event reasoning, an evolution feature extraction module, a dynamic emotion graph generation module and a stock variation prediction module, which may specifically be as follows:
(1) An index learning module based on event reasoning: a financial domain's theory map was constructed from the current dataset, which delineated the development laws of financial events, containing 17 common financial events. For the input news texts, the news texts are classified by adopting a trimmed BERT model, so that the news texts can be classified into one of 17 types of financial events. For the event represented by each news, the event represented by each news is caused to randomly walk on a map of a matter, the news after each classification starts from the node after the classification, and the event propagation value corresponding to each news text is obtained according to the weight of each side as the probability of transition until the event 'stock rising' or the event 'stock falling' is reached. The event propagation value of the news text of each day is averaged to represent the current news feature of the day, and the current influence of the historical data is represented by the EMW (a technical analysis index) value, so that the target news feature of each stock is obtained. In addition, the digital characteristic of the price can be converted into the index characteristic, so that the digital characteristic is obtained.
(2) The evolution characteristic extraction module is used for: and fusing the digital characteristics with the target news characteristics, so as to obtain the interactive characteristics of each stock in L days, extracting time sequence characteristics by adopting GRU, and taking the time sequence characteristics as the stock characteristics of the stock.
(3) A dynamic emotion figure generation module: and classifying the emotion tendencies of the comment texts through the BERT model to obtain two types of comment texts with positive emotion and comment texts with negative emotion. And counting the text quantity of comment texts of the positive emotion and the negative emotion in one day, so as to obtain emotion tendency scores of stocks in each day, and for the d day, adopting GRU network coding to the emotion tendency scores of the previous (d-L) day into a unified vector, so as to obtain the emotion characteristics of the stocks on the d day. And extracting the interrelationship of the comment emotion characteristics by adopting a two-layer one-dimensional convolutional neural network, thereby obtaining the current relation matrix between stocks. Considering that the past relationship should have a certain influence on the present, the last generated relationship matrix is added to the current relationship matrix, and the last relationship matrix is generated through an activation function, so as to obtain the target relationship matrix (feature) of each day.
(4) Stock variation prediction module: the target stock characteristics of the ith stock generated through m attention headers based on the target relation matrix and the stock characteristics of the stock through the GAT network are output by using a single-layer neural network, so that the trading trend of the ith stock is obtained.
The method of the scheme may further include various variants, for example, 3 indexes (the method does not perform indexing processing on the original data), no dynamic graph generation (the method does not dynamically generate graphs, but uses a fixed stock-to-stock relationship matrix), a splicing method (the method uses splicing operation during feature interaction), a linear layer method (the method uses a single-layer neural network during feature interaction), a single input (the method does not add event parts based on a rational graph), and the like. For this protocol, compared with the variant based on this protocol, ACC (a predicted task evaluation index) and AUC (a predicted task evaluation index) were selected as evaluation indexes, and the test results under 30 days of data can be shown in table 2:
TABLE 2
| Prediction method | ACC | AUC |
| 3 index | 0.5332 | 0.5618 |
| Non-dynamic graph generation | 0.5452 | 0.5483 |
| Splicing method | 0.5264 | 0.5444 |
| Linear layer process | 0.5331 | 0.5552 |
| Single input | 0.5489 | 0.5516 |
| The proposal is that | 0.5668 | 0.5662 |
Among them, it can be found that the prediction effect of the present scheme is better than that of the variants based on the present scheme, i.e. the prediction accuracy is highest.
In addition, other methods for predicting the trading trend of stocks can be selected for comparison, and the test results under 30 days of data can be shown in table 3:
TABLE 3 Table 3
The method has better prediction effect than other methods for predicting the trading trend of stocks, namely the prediction accuracy is highest.
The scheme combines the digital transaction data, the content text and the comment text of the transaction object, and provides a more comprehensive analysis basis. By comprehensively analyzing the quantities, the causal relationship, the emotion transfer effect and the dynamic association among the transaction objects in the transaction market can be captured, so that the prediction accuracy of the transaction trend of the transaction objects can be improved, and the accuracy of data processing is further improved.
When predicting the transaction trend of the transaction object, the overall operation mode may include collecting digital transaction data of the transaction object, and text data such as related content text and interactive text, and preprocessing and cleaning the text data, so as to obtain training data of the prediction model. The algorithm and the method provided by the scheme are used for model training and establishment by combining prepared training data. This includes event deduction and causal relationship analysis using a physiological map, and transaction object association analysis using a dynamic map to construct emotion maps and map neural networks. After model training is completed, the model can be used for stock price trend prediction by inputting stocks and related data. The model of the scheme analyzes the digital characteristics, content text, and interactive text of the transaction object and generates predictive results to assist the target object in making investment decisions.
Optionally, in some embodiments, the trading trend of each trading object may also be predicted based on a reinforcement learning prediction model that treats the trading market as a reinforcement learning environment, and learns optimal investment strategies based on historical data and market conditions by interacting with the environment. The status representation may be constructed using data such as transaction price and transaction amount of the historical transaction target. Through calculation of technical indexes and characteristic engineering, historical information of the trading market is converted into a state representation which is understandable by a model, so that trends and modes of the market can be captured better. Finally, decisions can also be made in real time based on the latest market conditions, without relying on predefined rules or specific data sources. This allows for more flexibility in coping with market changes and fluctuations.
As can be seen from the above, in the embodiment of the present application, after object data of at least one transaction object in multiple time periods is obtained, the object data includes digital transaction data and multiple text data, the text data includes content text and interactive text, a transaction event is extracted from the content text, and text reasoning is performed on the transaction event by using a preset event map to obtain content features of each transaction object, then digital features are extracted from the digital transaction data, and the content features and the digital features of the transaction object are fused to obtain object features of each transaction object, emotion features are extracted from the interactive text, and target relationship features between the transaction objects are determined according to feature relationships between emotion features of different transaction objects, then the target relationship features and the object features are spliced to obtain target object features of each transaction object, and transaction trends of each transaction object are predicted based on the target object features; according to the scheme, aiming at neutral content texts, transaction events are extracted from the content texts, and text reasoning is carried out on the transaction events by adopting a preset rational map, so that causal relations in a transaction market are captured, more reliable logic information is provided for predicting transaction areas of transaction objects, interactive texts are introduced, emotion features are extracted from the interactive texts, and the current relation among the transaction objects is determined based on the emotion features, so that the transfer effect of emotion of the transaction objects or the transaction market can be captured, and further, the accuracy of predicting transaction trends is improved, and therefore, the accuracy of data processing can be improved.
According to the method described in the above embodiments, examples are described in further detail below.
In this embodiment, the data processing apparatus is specifically integrated in an electronic device, the electronic device is a server, the transaction object is a stock, the time period is 1 day, the content text is a news text, and the interactive text is a comment text.
As shown in fig. 5, a data processing method specifically includes the following steps:
201. the server obtains object data for at least one stock over a plurality of days.
For example, the server may directly receive object data of at least one stock on multiple days sent by the terminal, or may extract current trade data and history trade data of at least one stock in a trade platform or a trade database, extract digital trade data of multiple days in the current trade data and the history trade data, extract multiple news texts and multiple comment texts of at least one stock on multiple days in a network or a text database, thereby obtaining text data of the stock, use the text data and the digital trade data as object data of the stock on multiple days, or, when the number of the object data is large or the occupied memory is large, may also receive a data processing request, where the data processing request carries a storage address of the object data of at least one stock on multiple days, obtain the object data of at least one stock on multiple days based on the storage address, and so on.
202. The server extracts transaction events from the news text.
For example, the server may obtain a historical transaction data set in a preset historical time range, extract target transaction data corresponding to a plurality of preset transaction times from the historical transaction data set, determine a logical relationship between preset events based on the target transaction data, construct a rational graph with the transaction events as nodes based on the logical relationship, and use the rational graph as the preset rational graph.
The server adopts the trimmed BERT to perform feature coding on the news text to obtain the text feature of the news text, or can also adopt other coding networks capable of performing content coding to perform feature coding on the news text to obtain the text feature of the news text, and the like.
The server may match the text feature with the event feature corresponding to each transaction event type, and use the transaction event type corresponding to the successfully matched event feature as the transaction event type corresponding to the news text, or may also obtain the event feature corresponding to each transaction event, calculate the feature similarity or feature distance between the text feature and the event feature, and determine the transaction event type corresponding to the news text based on the feature similarity or feature distance, or classify the news text based on the text feature by using a classification network, to obtain the transaction event type corresponding to the news text, and so on.
The server screens out preset transaction events corresponding to the transaction event types from the preset transaction events to obtain transaction events corresponding to the news text.
203. The server adopts a preset event map to carry out text reasoning on the transaction event so as to obtain the news characteristic of each stock.
For example, the server may identify, in a preset event map, a weight corresponding to a node edge between nodes, and take the weight as a probability of a directed transition between the corresponding two nodes. And screening target event nodes corresponding to the transaction event from the event nodes based on the event type of the transaction event.
The server may randomly walk the transaction event from the target event node to the termination node in a preset event map according to the transition probability, or may determine at least one transition path of the target event node to the termination node in the preset event map according to the transition probability, and based on the transition path, shift or walk the transaction event from the target event node in the preset event map. When the transaction event reaches any one of the termination nodes, stopping transferring the transaction event, and taking the reached termination node as a target termination node. The event type corresponding to the target termination node is obtained, and a preset event propagation value corresponding to the event type is screened out from a preset event propagation value set to obtain an event propagation value of the news text.
The server can screen out the time propagation value of the news text of the same day from the time propagation values, so as to obtain a time propagation value set corresponding to each day, and calculate the average value of the time propagation values in the time propagation value set, so as to obtain the initial news characteristic corresponding to each day.
The server may randomly or arbitrarily screen out one day as the target time period in a plurality of days, or may screen out the target time period in a plurality of days based on timing information of a plurality of days, or the like.
The server screens out target initial news features of a target time period from the initial news features. Based on the time sequence information of a plurality of days, at least one historical time period corresponding to the target time period is screened out in the plurality of days, initial news characteristics corresponding to the historical time period are extracted from the initial news characteristics, the historical news characteristics are obtained, and data weights corresponding to the target initial news characteristics are obtained. Based on the data weight, determining initial adjustment weight corresponding to the historical news feature, calculating a time distance between each historical time period and a target time period to obtain a target time distance corresponding to each historical time period, and fusing the target time distance and the initial adjustment weight to obtain the adjustment weight corresponding to the historical news feature. And weighting the historical news characteristics according to the adjustment weight.
The server can accumulate the weighted historical news features and the target initial news features to obtain fused news features, fuse the adjustment weights and preset fusion weights corresponding to the target initial news features to obtain fused weights, and calculate the ratio between the fused news features and the fused weights to obtain the current news features of the target time period. Taking the target time period as the d day as an example, the specific formula can be shown as formula (1).
After the server adjusts the target initial news feature, the step of determining the target time period in a plurality of days can be carried out again until each day in the plurality of days is the target time period, so that the current news feature of each day is obtained, the current news feature of each day is used as the news feature of the stock, and the news feature of each stock is obtained.
204. The server extracts digital features from the digital transaction data.
For example, the server may perform feature encoding on the digital transaction data to obtain digital sub-features of each day, and use the digital sub-features as digital features, or may extract digital data corresponding to the target transaction index from the digital transaction data, and perform feature encoding or conversion on the digital data, so as to obtain digital sub-features of each day, use the digital sub-features as digital features, and so on.
205. And the server fuses the news characteristics and the digital characteristics of the stocks to obtain the object characteristics of each stock.
For example, the server may screen out the target news feature corresponding to the digital sub-feature from the news features based on the time period corresponding to the digital sub-feature. The digital sub-features of each day are spliced with the target news features to obtain spliced object features of each day, the digital sub-features of each day are multiplied with the target news features to obtain candidate object features of each day, the candidate object features are fused with the spliced object features to obtain corresponding interaction features of each day, and then interaction features of each object in multiple days are obtained, which can be shown as a formula (2). The GRU is adopted to extract the time sequence feature from the interaction feature, and the time sequence feature is used as the object feature of the stock, which can be shown as a formula (3). After the time sequence feature is extracted from the interaction feature, the time sequence feature can be used as the object feature of the stock.
206. And the server extracts emotion characteristics from the comment text.
For example, the server may use the BERT model to perform emotion classification on the comment text, so as to obtain an emotion type of the comment text, or may use other classification models (networks) that may perform emotion classification to perform emotion classification on the comment text, so as to obtain an emotion type of the comment text, or the like.
The method comprises the steps that a server screens comment texts of the same emotion type in each day from comment texts to obtain a positive comment text set corresponding to positive emotion of each day and a negative comment text set corresponding to negative emotion, the number of the comment texts is counted in the positive comment text set to obtain the number of the positive texts of each day, and the number of the comment texts is counted in the negative comment text set to obtain the number of the negative texts of each day.
The server may determine positive emotion parameters of each day based on the number of positive texts, determine negative emotion parameters of each day according to the number of negative texts, and calculate a ratio between the positive emotion parameters and the negative emotion parameters of the same day to obtain emotion tendency parameters corresponding to stocks in each day, which may be shown in formula (4).
The server identifies at least one historical time period corresponding to each day in the time period to obtain a target historical time period of each day, and screens out emotion tendency parameters corresponding to the target historical time period from the emotion tendency parameters to obtain the historical emotion tendency parameters of each day. And (3) adopting the GRU network to encode the emotion tendency parameters and the historical emotion tendency parameters of each day into a unified vector, and taking the unified vector as the emotion characteristics of the day.
207. And the server determines target relation characteristics among the stocks according to the characteristic relation among the emotion characteristics of different stocks.
For example, the server may screen the emotion features of different stocks in each day from the emotion features to obtain an emotion feature set corresponding to each day.
The server may adopt a broadcast operation mechanism to perform broadcast addition on the emotion features in the emotion feature set to obtain fused emotion features, or may also adopt an attention network to extract associated features of the emotion features in the emotion feature set, determine attention weights of the emotion features based on the associated features, weight the emotion features based on the attention weights, and fuse the weighted emotion features to obtain fused emotion features.
The server can adopt a one-dimensional convolution neural network of a first layer to extract the relation features from the fused emotion features, and adopts a one-dimensional convolution neural network of a second layer to perform feature conversion on the relation features so as to obtain the current relation features between stocks of each day, an activation function corresponding to the one-dimensional convolution neural network of the two layers can be ELU, and the relation features between the emotion features extracted from the emotion feature set can be shown as a formula (6).
The server acquires the history relation characteristic among stocks in a preset history time period, when the preset history time period is not in a plurality of time periods, the time period with the longest time distance from the current moment is screened out of the time periods to serve as a candidate time period adjacent to the preset history time period, and when the preset history time is in the plurality of time periods (namely, the preset history time period is one day in the plurality of time periods), the time period adjacent to the preset history time period can be screened out of the time periods, so that the candidate time period is obtained. And adding the current relation feature corresponding to the candidate time period and the history relation feature to obtain a fusion relation feature, and mapping the fusion relation feature to the target relation feature through an activation function, wherein the fusion relation feature is specifically shown in a formula (7).
The server may take the candidate time period as a preset historical time period, take the target relationship feature as a historical relationship feature of the preset historical time period, and then return to execute the step of screening out the candidate time period adjacent to the preset historical time period in the time period until the time periods are all candidate time periods, so as to obtain the target relationship feature corresponding to each day.
208. And the server splices the target relation characteristic and the object characteristic to obtain the target object characteristic of each stock.
For example, the server may generate m candidate object features based on the target relationship features and the object features using m attention headers in the GAT network, and splice the m candidate object features to obtain target object features of the stock on each day, which may be shown in formula (8).
209. Based on the target object characteristics, a trading trend of each stock is predicted.
For example, the server predicts the trading trend of each stock based on the target object characteristics using a single-layer prediction network, so as to obtain a predicted trading trend of each object, which may be shown in equation (9).
As can be seen from the foregoing, after obtaining object data of at least one stock on multiple days, the server in this embodiment includes digital transaction data and multiple text data, the text data includes news text and comment text, a transaction event is extracted from the news text, and text reasoning is performed on the transaction event by using a preset rational map, so as to obtain news features of each stock, then digital features are extracted from the digital transaction data, and the news features and the digital features of the stock are fused to obtain object features of each stock, emotion features are extracted from the comment text, and target relationship features between the stocks are determined according to feature relationships between emotion features of different stocks, then the target relationship features and the object features are spliced to obtain target object features of each stock, and transaction trends of each stock are predicted based on the target object features; according to the scheme, aiming at neutral news texts, transaction events are extracted from the news texts, and text reasoning is carried out on the transaction events by adopting a preset rational map, so that causal relations in a transaction market are captured, more reliable logic information is provided for predicting a transaction area of stocks, comment texts are introduced, emotion features are extracted from the comment texts, and the current relation among stocks is determined based on emotion feature structures, so that the transfer effect of stocks or the emotion of the transaction market can be captured, and further the accuracy of transaction trend prediction is improved, and therefore, the accuracy of data processing can be improved.
In order to better implement the above method, the embodiment of the present invention further provides a data processing apparatus, where the data processing apparatus may be integrated into an electronic device, such as a server or a terminal, where the terminal may include a tablet computer, a notebook computer, and/or a personal computer.
For example, as shown in fig. 6, the data processing apparatus may include an acquisition unit 301, an inference unit 302, a fusion unit 303, a determination unit 304, and a prediction unit 305, as follows:
(1) An acquisition unit 301;
an obtaining unit 301 is configured to obtain object data of at least one transaction object in a plurality of time periods, where the object data includes digital transaction data and a plurality of text data, and the text data includes content text and interactive text.
For example, the obtaining unit 301 may specifically be configured to directly receive object data of at least one transaction object sent by a terminal in multiple time periods, or extract current transaction data and historical transaction data of the at least one transaction object in a transaction platform or a transaction database, extract digital transaction data of the at least one transaction object in multiple time periods in the current transaction data and the historical transaction data respectively, extract multiple content texts and multiple interactive texts of the at least one transaction object in multiple time periods in a network or a text database, thereby obtaining text data of the transaction object, and use the text data and the digital transaction data as object data of the transaction object in multiple time periods, or when the number of the object data is large or occupied by a memory is large, receive a data processing request, where the data processing request carries a storage address of the object data of the at least one transaction object in multiple time periods, obtain the object data of the at least one transaction object in multiple time periods based on the storage address, and so on.
(2) An inference unit 302;
the reasoning unit 302 is configured to extract a transaction event from the content text, and perform text reasoning on the transaction event by using a preset event map, so as to obtain content features of each transaction object.
For example, the inference unit 302 may be specifically configured to extract a transaction event from a content text, propagate the transaction event in a preset event map to obtain an event propagation value of each content text, count a mean value of event propagation values of each time period in an event propagation statistics value to obtain an initial content feature of a transaction object in each time period, and adjust the initial content feature based on time sequence information of the time period to obtain a content feature of each transaction object.
(3) A fusion unit 303;
and a fusion unit 303, configured to extract digital features from the digital transaction data, and fuse the content features and the digital features of the transaction objects to obtain object features of each transaction object.
For example, the fusion unit 303 may specifically be configured to extract a digital feature from digital transaction data, screen a target content feature corresponding to the digital sub-feature from the content features based on a time period corresponding to the digital sub-feature, splice the digital sub-feature and the target content feature in each time period to obtain an interaction feature of each transaction object in a plurality of time periods, extract a time sequence feature from the interaction feature, and use the time sequence feature as an object feature of the transaction object.
(4) A determination unit 304;
and the determining unit 304 is configured to extract emotion features from the interactive text, and determine target relationship features between the transaction objects according to feature relationships between emotion features of different transaction objects.
For example, the determining unit 304 may be specifically configured to extract emotion features from the interactive text, screen emotion features of different transaction objects in each time period from the emotion features, obtain an emotion feature set corresponding to each time period, extract relationship features between emotion features from the emotion feature set, so as to obtain a current relationship feature between transaction objects in each time period, and adjust the current relationship feature based on time sequence information of the time period, so as to obtain a target relationship feature corresponding to each time period.
(5) A prediction unit 305;
the prediction unit 305 is configured to splice the target relationship feature and the object feature to obtain a target object feature of each transaction object, and predict a transaction trend of each transaction object based on the target object feature.
For example, the prediction unit 305 may be specifically configured to determine the attention weight of each transaction object in multiple dimensions based on the target relationship feature, weight the object feature of each transaction object according to the attention weight to obtain candidate object features of each transaction object in multiple dimensions, splice the candidate object features to obtain the target object feature of each transaction object, and predict the transaction trend of each transaction object based on the target object feature.
In the implementation, each unit may be implemented as an independent entity, or may be implemented as the same entity or several entities in any combination, and the implementation of each unit may be referred to the foregoing method embodiment, which is not described herein again.
As can be seen from the foregoing, in this embodiment, after the obtaining unit 301 obtains object data of at least one transaction object in a plurality of time periods, the object data includes digital transaction data and a plurality of text data, the text data includes content text and interactive text, the inference unit 302 extracts transaction events in the content text, and performs text inference on the transaction events by using a preset rational map to obtain content features of each transaction object, then the fusion unit 303 extracts digital features from the digital transaction data, fuses the content features and the digital features of the transaction object to obtain object features of each transaction object, the determination unit 304 extracts emotion features in the interactive text, and determines target relationship features between the transaction objects according to feature relationships between emotion features of different transaction objects, and then the prediction unit 305 splices the target relationship features with the object features to obtain target object features of each transaction object, and predicts transaction trends of each transaction object based on the target object features; according to the scheme, aiming at neutral content texts, transaction events are extracted from the content texts, and text reasoning is carried out on the transaction events by adopting a preset rational map, so that causal relations in a transaction market are captured, more reliable logic information is provided for predicting transaction areas of transaction objects, interactive texts are introduced, emotion features are extracted from the interactive texts, and the current relation among the transaction objects is determined based on the emotion features, so that the transfer effect of emotion of the transaction objects or the transaction market can be captured, and further, the accuracy of predicting transaction trends is improved, and therefore, the accuracy of data processing can be improved.
The embodiment of the invention also provides an electronic device, as shown in fig. 7, which shows a schematic structural diagram of the electronic device according to the embodiment of the invention, specifically:
the electronic device may include one or more processing cores 'processors 401, one or more computer-readable storage media's memory 402, power supply 403, and input unit 404, among other components. It will be appreciated by those skilled in the art that the electronic device structure shown in fig. 7 is not limiting of the electronic device and may include more or fewer components than shown, or may combine certain components, or a different arrangement of components. Wherein:
the processor 401 is a control center of the electronic device, connects various parts of the entire electronic device using various interfaces and lines, and performs various functions of the electronic device and processes data by running or executing software programs and/or modules stored in the memory 402, and calling data stored in the memory 402. Optionally, processor 401 may include one or more processing cores; preferably, the processor 401 may integrate an application processor and a modem processor, wherein the application processor mainly processes an operating system, a user interface, an application program, etc., and the modem processor mainly processes wireless communication. It will be appreciated that the modem processor described above may not be integrated into the processor 401.
The memory 402 may be used to store software programs and modules, and the processor 401 executes various functional applications and data processing by executing the software programs and modules stored in the memory 402. The memory 402 may mainly include a storage program area and a storage data area, wherein the storage program area may store an operating system, an application program (such as a sound playing function, an image playing function, etc.) required for at least one function, and the like; the storage data area may store data created according to the use of the electronic device, etc. In addition, memory 402 may include high-speed random access memory, and may also include non-volatile memory, such as at least one magnetic disk storage device, flash memory device, or other volatile solid-state storage device. Accordingly, the memory 402 may also include a memory controller to provide the processor 401 with access to the memory 402.
The electronic device further comprises a power supply 403 for supplying power to the various components, preferably the power supply 403 may be logically connected to the processor 401 by a power management system, so that functions of managing charging, discharging, and power consumption are performed by the power management system. The power supply 403 may also include one or more of any of a direct current or alternating current power supply, a recharging system, a power failure detection circuit, a power converter or inverter, a power status indicator, and the like.
The electronic device may further comprise an input unit 404, which input unit 404 may be used for receiving input digital or character information and generating keyboard, mouse, joystick, optical or trackball signal inputs in connection with user settings and function control.
Although not shown, the electronic device may further include a display unit or the like, which is not described herein. In particular, in this embodiment, the processor 401 in the electronic device loads executable files corresponding to the processes of one or more application programs into the memory 402 according to the following instructions, and the processor 401 executes the application programs stored in the memory 402, so as to implement various functions as follows:
obtaining object data of at least one transaction object in a plurality of time periods, wherein the object data comprises digital transaction data and a plurality of text data, the text data comprises content text and interactive text, transaction events are extracted from the content text, text reasoning is carried out on the transaction events by adopting a preset event map so as to obtain content characteristics of each transaction object, digital characteristics are extracted from the digital transaction data, the content characteristics and the digital characteristics of each transaction object are fused to obtain object characteristics of each transaction object, emotion characteristics are extracted from the interactive text, target relation characteristics among the transaction objects are determined according to characteristic relations among the emotion characteristics of different transaction objects, the target relation characteristics are spliced with the object characteristics to obtain target object characteristics of each transaction object, and transaction trend of each transaction object is predicted based on the target object characteristics.
For example, the electronic device may obtain digital transaction data, a plurality of content texts and a plurality of interaction texts of at least one transaction object in a plurality of time periods, extract transaction events in the content texts, propagate the transaction events in a preset event map to obtain event propagation values of each content text, count a mean value of the event propagation values of each time period in the event propagation statistics values to obtain initial content features of the transaction object in each time period, and adjust the initial content features based on time sequence information of the time periods to obtain content features of each transaction object. Extracting digital characteristics from digital transaction data, screening target content characteristics corresponding to the digital sub-characteristics from the content characteristics based on time periods corresponding to the digital sub-characteristics, splicing the digital sub-characteristics of each time period with the target content characteristics to obtain interactive characteristics of each transaction object in a plurality of time periods, extracting time sequence characteristics from the interactive characteristics, and taking the time sequence characteristics as object characteristics of the transaction objects. Extracting emotion characteristics from the interactive text, screening emotion characteristics of different transaction objects in each time period from the emotion characteristics, obtaining emotion characteristic sets corresponding to each time period, extracting relation characteristics among the emotion characteristics from the emotion characteristic sets, obtaining current relation characteristics among the transaction objects in each time period, and adjusting the current relation characteristics based on time sequence information of the time period to obtain target relation characteristics corresponding to each time period. Based on the target relation characteristics, determining the attention weight of each transaction object in a plurality of dimensions, weighting the object characteristics of each transaction object according to the attention weight to obtain candidate object characteristics of each transaction object in the plurality of dimensions, splicing the candidate object characteristics to obtain target object characteristics of each transaction object, and predicting the transaction trend of each transaction object based on the target object characteristics.
The specific implementation of each operation may be referred to the previous embodiments, and will not be described herein.
As can be seen from the above, in the embodiment of the present invention, after object data of at least one transaction object in multiple time periods is obtained, the object data includes digital transaction data and multiple text data, the text data includes content text and interactive text, a transaction event is extracted from the content text, and text reasoning is performed on the transaction event by using a preset event map to obtain content features of each transaction object, then digital features are extracted from the digital transaction data, and the content features and the digital features of the transaction object are fused to obtain object features of each transaction object, emotion features are extracted from the interactive text, and target relationship features between the transaction objects are determined according to feature relationships between emotion features of different transaction objects, then the target relationship features and the object features are spliced to obtain target object features of each transaction object, and transaction trends of each transaction object are predicted based on the target object features; according to the scheme, aiming at neutral content texts, transaction events are extracted from the content texts, and text reasoning is carried out on the transaction events by adopting a preset rational map, so that causal relations in a transaction market are captured, more reliable logic information is provided for predicting transaction areas of transaction objects, interactive texts are introduced, emotion features are extracted from the interactive texts, and the current relation among the transaction objects is determined based on the emotion features, so that the transfer effect of emotion of the transaction objects or the transaction market can be captured, and further, the accuracy of predicting transaction trends is improved, and therefore, the accuracy of data processing can be improved.
Those of ordinary skill in the art will appreciate that all or a portion of the steps of the various methods of the above embodiments may be performed by instructions, or by instructions controlling associated hardware, which may be stored in a computer-readable storage medium and loaded and executed by a processor.
To this end, an embodiment of the present invention provides a computer readable storage medium having stored therein a plurality of instructions capable of being loaded by a processor to perform the steps of any of the data processing methods provided by the embodiments of the present invention. For example, the instructions may perform the steps of:
obtaining object data of at least one transaction object in a plurality of time periods, wherein the object data comprises digital transaction data and a plurality of text data, the text data comprises content text and interactive text, transaction events are extracted from the content text, text reasoning is carried out on the transaction events by adopting a preset event map so as to obtain content characteristics of each transaction object, digital characteristics are extracted from the digital transaction data, the content characteristics and the digital characteristics of each transaction object are fused to obtain object characteristics of each transaction object, emotion characteristics are extracted from the interactive text, target relation characteristics among the transaction objects are determined according to characteristic relations among the emotion characteristics of different transaction objects, the target relation characteristics are spliced with the object characteristics to obtain target object characteristics of each transaction object, and transaction trend of each transaction object is predicted based on the target object characteristics.
For example, the electronic device may obtain digital transaction data, a plurality of content texts and a plurality of interaction texts of at least one transaction object in a plurality of time periods, extract transaction events in the content texts, propagate the transaction events in a preset event map to obtain event propagation values of each content text, count a mean value of the event propagation values of each time period in the event propagation statistics values to obtain initial content features of the transaction object in each time period, and adjust the initial content features based on time sequence information of the time periods to obtain content features of each transaction object. Extracting digital characteristics from digital transaction data, screening target content characteristics corresponding to the digital sub-characteristics from the content characteristics based on time periods corresponding to the digital sub-characteristics, splicing the digital sub-characteristics of each time period with the target content characteristics to obtain interactive characteristics of each transaction object in a plurality of time periods, extracting time sequence characteristics from the interactive characteristics, and taking the time sequence characteristics as object characteristics of the transaction objects. Extracting emotion characteristics from the interactive text, screening emotion characteristics of different transaction objects in each time period from the emotion characteristics, obtaining emotion characteristic sets corresponding to each time period, extracting relation characteristics among the emotion characteristics from the emotion characteristic sets, obtaining current relation characteristics among the transaction objects in each time period, and adjusting the current relation characteristics based on time sequence information of the time period to obtain target relation characteristics corresponding to each time period. Based on the target relation characteristics, determining the attention weight of each transaction object in a plurality of dimensions, weighting the object characteristics of each transaction object according to the attention weight to obtain candidate object characteristics of each transaction object in the plurality of dimensions, splicing the candidate object characteristics to obtain target object characteristics of each transaction object, and predicting the transaction trend of each transaction object based on the target object characteristics.
The specific implementation of each operation above may be referred to the previous embodiments, and will not be described herein.
Wherein the computer-readable storage medium may comprise: read Only Memory (ROM), random access Memory (RAM, random Access Memory), magnetic or optical disk, and the like.
Because the instructions stored in the computer readable storage medium may execute the steps in any data processing method provided by the embodiments of the present application, the beneficial effects that any data processing method provided by the embodiments of the present application can be achieved, which are detailed in the previous embodiments and are not described herein.
Wherein according to an aspect of the application, a computer program product or a computer program is provided, the computer program product or computer program comprising computer instructions stored in a computer readable storage medium. The computer instructions are read from a computer-readable storage medium by a processor of an electronic device, and executed by the processor, cause the electronic device to perform the methods provided in various alternative implementations of the data processing aspects or transaction trend prediction aspects of a transaction object described above.
The foregoing has described in detail a data processing method, apparatus, electronic device and computer readable storage medium according to embodiments of the present invention, and specific examples have been applied to illustrate the principles and embodiments of the present invention, where the foregoing examples are provided to assist in understanding the method and core idea of the present invention; meanwhile, as those skilled in the art will have variations in the specific embodiments and application scope in light of the ideas of the present invention, the present description should not be construed as limiting the present invention.
Claims (20)
1. A method of data processing, comprising:
acquiring object data of at least one transaction object in a plurality of time periods, wherein the object data comprises digital transaction data and a plurality of text data, and the text data comprises content text and interactive text;
extracting transaction events from the content text, and carrying out text reasoning on the transaction events by adopting a preset rational map so as to obtain the content characteristics of each transaction object;
extracting digital characteristics from the digital transaction data, and fusing the content characteristics and the digital characteristics of the transaction objects to obtain object characteristics of each transaction object;
Extracting emotion characteristics from the interactive text, and determining target relationship characteristics among the transaction objects according to characteristic relationships among the emotion characteristics of different transaction objects;
and splicing the target relation characteristic with the object characteristic to obtain a target object characteristic of each transaction object, and predicting the transaction trend of each transaction object based on the target object characteristic.
2. The data processing method according to claim 1, wherein the text reasoning is performed on the transaction events using a preset event map to obtain content characteristics of each transaction object, including:
propagating the transaction event in a preset event map to obtain an event propagation value of each content text, wherein the event propagation value represents a logical relationship between the transaction event and a transaction trend;
calculating the average value of the event propagation values of each time period in the event propagation values to obtain the initial content characteristics of the transaction object in each time period;
and adjusting the initial content characteristics based on the time sequence information of the time period to obtain the content characteristics of each transaction object.
3. The method according to claim 2, wherein the predetermined event map includes a plurality of nodes having a logical relationship, the nodes include event nodes, and the step of propagating the transaction event in the predetermined event map to obtain an event propagation value for each content text includes:
identifying transition probabilities among the nodes in a preset event map;
based on the event type of the transaction event, selecting a target event node corresponding to the transaction event from the event nodes;
and according to the transition probability, the transaction event is transferred from the target event node in the preset event map so as to obtain an event propagation value of each content text.
4. A data processing method according to claim 3, wherein the nodes further comprise a plurality of termination nodes, the termination nodes comprising nodes indicating transaction trends, the transferring the transaction event from the target event node in the preset event map according to the transfer probabilities to obtain event propagation values for each content text, comprising:
according to the transition probability, the transaction event is transferred from the target event node to the termination node in the preset event map;
Stopping transferring the transaction event when the transaction event reaches any one termination node, and taking the reached termination node as a target termination node;
and determining an event propagation value of the content text based on the event type corresponding to the target termination node.
5. The data processing method according to claim 2, wherein the adjusting the initial content feature based on the time-series information of the time period to obtain the content feature of each transaction object includes:
determining a target time period in the time periods, and screening out target initial content characteristics of the target time period from the initial content characteristics;
adjusting the target initial content characteristics based on the time sequence information of the time period to obtain the current content characteristics of the target time period;
returning to the step of determining the target time period in the time period until the time period is the target time period, and obtaining the current content characteristics of each time period;
and taking the current content characteristic of each time period as the content characteristic of the transaction object.
6. The method according to claim 5, wherein the adjusting the target initial content feature based on the time sequence information of the time period to obtain the current content feature of the target time period includes:
screening at least one historical time period corresponding to the target time period from the time periods based on the time sequence information of the time periods;
extracting initial content characteristics corresponding to the historical time period from the initial content characteristics to obtain historical content characteristics;
acquiring data weight corresponding to the target initial content characteristics, and determining adjustment weight corresponding to the historical content characteristics based on the data weight;
and weighting the historical content characteristics according to the adjustment weight, and fusing the weighted historical content characteristics with the target initial content characteristics to obtain the current content characteristics of the target time period.
7. The method of claim 2, wherein before the step of propagating the transaction event in a preset event map to obtain an event propagation value of each content text, further comprises:
Acquiring a historical transaction data set of a preset historical time range;
extracting target transaction data corresponding to a plurality of preset transaction events from the historical transaction data set, and determining a logic relationship between the preset transaction events based on the target transaction data;
based on the logical relationship, constructing a rational map by taking the preset transaction event as a node, and taking the rational map as a preset rational map.
8. The data processing method of claim 7, wherein extracting transaction events in the content text comprises:
extracting features of the content text to obtain text features of the content text;
classifying the content text based on the text characteristics to obtain transaction event types corresponding to the content text;
and screening out a preset transaction event corresponding to the transaction event type from the preset transaction event to obtain a transaction event corresponding to the content text.
9. The method according to any one of claims 1 to 8, wherein the digital features include digital sub-features for each time period, and the fusing the content features and the digital features of the transaction objects to obtain object features of each transaction object includes:
Screening out target content features corresponding to the digital sub-features from the content features based on the time periods corresponding to the digital sub-features;
splicing the digital sub-features and the target content features of each time period to obtain interaction features of each transaction object in a plurality of time periods;
and extracting time sequence features from the interaction features, and taking the time sequence features as object features of the transaction objects.
10. The method according to any one of claims 1 to 8, wherein extracting emotion features from the interactive text comprises:
carrying out emotion classification on the interactive texts, and counting the number of texts of different emotion types in each time period;
determining emotion tendency parameters corresponding to the transaction objects in each time period based on the text quantity, wherein the emotion tendency parameters indicate emotion tendency degrees of interaction aiming at the interaction objects;
and carrying out feature coding on the emotion tendency parameters to obtain emotion features of the transaction object in each time period.
11. The data processing method according to claim 10, wherein the text quantity includes a positive text quantity and a negative text quantity, and the determining, based on the text quantity, emotion tendency parameters corresponding to the transaction object in each time period includes:
Determining positive emotion parameters of each time period based on the number of positive texts;
according to the number of the negative texts, determining the negative emotion parameters of each time period;
and calculating the ratio between the positive emotion parameters and the negative emotion parameters in the same time period to obtain emotion tendency parameters corresponding to the transaction objects in each time period.
12. The method of claim 10, wherein the feature encoding the emotion tendencies parameters to obtain emotion characteristics for the transaction object for each time period comprises:
identifying at least one historical time period corresponding to each time period in the time periods to obtain a target historical time period of each time period;
screening emotion tendency parameters corresponding to the target historical time period from the emotion tendency parameters to obtain historical emotion tendency parameters of each time period;
and carrying out feature coding on the emotion tendency parameters and the historical emotion tendency parameters of each time period to obtain emotion features of each time period.
13. The method according to any one of claims 1 to 8, wherein the determining the target relationship feature between the transaction objects according to the feature relationship between the emotion features of the different transaction objects includes:
Screening emotion characteristics of different transaction objects in each time period from the emotion characteristics to obtain emotion characteristic sets corresponding to each time period;
extracting relation features among emotion features from the emotion feature set to obtain current relation features among the transaction objects in each time period;
and adjusting the current relation characteristic based on the time sequence information of the time period to obtain a target relation characteristic corresponding to each time period.
14. The method of claim 13, wherein extracting the relationship features between emotion features from the set of emotion features to obtain the current relationship features between the transaction objects for each time period comprises:
fusing the emotion characteristics in the emotion characteristic set to obtain fused emotion characteristics;
extracting relation features from the fused emotion features, wherein the relation features represent correlations among the emotion features in the emotion feature set;
and performing feature conversion on the relation features to obtain the current relation features among the transaction objects in each time period.
15. The method for processing data according to claim 13, wherein the adjusting the current relationship feature based on the time sequence information of the time period to obtain the target relationship feature corresponding to each time period includes:
acquiring historical relation characteristics among the transaction objects in a preset historical time period, and screening candidate time periods adjacent to the preset historical time period from the time periods;
fusing the current relation characteristic corresponding to the candidate time period with the history relation characteristic to obtain a target relation characteristic corresponding to the candidate time period;
taking the candidate time period as the preset historical time period, and taking the target relation characteristic as the historical relation characteristic of the preset historical time period;
and returning to the step of screening out the candidate time periods adjacent to the preset historical time period in the time periods until the time periods are all the candidate time periods, and obtaining the target relation characteristic corresponding to each time period.
16. The method according to any one of claims 1 to 8, wherein the stitching the target relationship feature with the object feature to obtain a target object feature of each transaction object includes:
Determining the attention weight of each transaction object in a plurality of dimensions based on the target relation features;
weighting the object characteristics of each transaction object according to the attention weight to obtain candidate object characteristics of each pair of transaction objects in multiple dimensions;
and splicing the candidate object features to obtain the target object features of each transaction object.
17. A data processing apparatus, comprising:
an acquisition unit, configured to acquire object data of at least one transaction object in a plurality of time periods, where the object data includes digital transaction data and a plurality of text data, and the text data includes content text and interactive text;
the reasoning unit is used for extracting transaction events from the content text and carrying out text reasoning on the transaction events by adopting a preset event map so as to obtain the content characteristics of each transaction object;
the fusion unit is used for extracting digital characteristics from the digital transaction data, and fusing the content characteristics and the digital characteristics of the transaction objects to obtain object characteristics of each transaction object;
the determining unit is used for extracting emotion characteristics from the interactive text and determining target relationship characteristics among the transaction objects according to characteristic relationships among the emotion characteristics of different transaction objects;
And the prediction unit is used for splicing the target relation characteristic with the object characteristic to obtain a target object characteristic of each transaction object, and predicting the transaction trend of each transaction object based on the target object characteristic.
18. An electronic device comprising a processor and a memory, the memory storing an application, the processor being configured to run the application in the memory to perform the steps in the data processing method of any of claims 1 to 16.
19. A computer program product comprising computer programs/instructions which, when executed by a processor, implement the steps of the data processing method of any of claims 1 to 16.
20. A computer readable storage medium storing a plurality of instructions adapted to be loaded by a processor to perform the steps of the data processing method of any of claims 1 to 16.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310900545.7A CN116975290A (en) | 2023-07-21 | 2023-07-21 | Data processing method, device, electronic equipment and computer readable storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310900545.7A CN116975290A (en) | 2023-07-21 | 2023-07-21 | Data processing method, device, electronic equipment and computer readable storage medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116975290A true CN116975290A (en) | 2023-10-31 |
Family
ID=88482538
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310900545.7A Pending CN116975290A (en) | 2023-07-21 | 2023-07-21 | Data processing method, device, electronic equipment and computer readable storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116975290A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN118153684A (en) * | 2024-05-09 | 2024-06-07 | 中国科学技术大学 | Knowledge-driven large model emotion tracing and propagation path analysis method |
-
2023
- 2023-07-21 CN CN202310900545.7A patent/CN116975290A/en active Pending
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN118153684A (en) * | 2024-05-09 | 2024-06-07 | 中国科学技术大学 | Knowledge-driven large model emotion tracing and propagation path analysis method |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113822494B (en) | Risk prediction method, device, equipment and storage medium | |
| Milana et al. | Artificial intelligence techniques in finance and financial markets: a survey of the literature | |
| WO2022161202A1 (en) | Multimedia resource classification model training method and multimedia resource recommendation method | |
| US20210264272A1 (en) | Training method and system of neural network model and prediction method and system | |
| WO2020249125A1 (en) | Method and system for automatically training machine learning model | |
| US10387536B2 (en) | Computerized data-aware agent systems for retrieving data to serve a dialog between human user and computerized system | |
| CN109766454A (en) | An investor classification method, device, equipment and medium | |
| CN116595328B (en) | Knowledge-graph-based intelligent construction device and method for data scoring card model | |
| Le et al. | A multi-criteria collaborative filtering approach using deep learning and Dempster-Shafer theory for hotel recommendations | |
| CN114819967A (en) | Data processing method and device, electronic equipment and computer readable storage medium | |
| CN116975290A (en) | Data processing method, device, electronic equipment and computer readable storage medium | |
| Wang et al. | A novel stock index direction prediction based on dual classifier coupling and investor sentiment analysis | |
| Zhu et al. | Intelligent product redesign strategy with ontology-based fine-grained sentiment analysis | |
| Voronov et al. | Forecasting popularity of news article by title analyzing with BN-LSTM network | |
| Nayak et al. | Feasibility study of machine learning & AI algorithms for classifying software requirements | |
| Drinkall et al. | Forecasting Credit Ratings: A Case Study where Traditional Methods Outperform Generative LLMs | |
| CN116186388A (en) | User's interest recommendation method, electronic device and storage medium | |
| CN115439180A (en) | Target object determination method and device, electronic equipment and storage medium | |
| Hirano et al. | STBM+: Advanced Stochastic Trading Behavior Model for Financial Markets using Residual Blocks or Transformers | |
| US12038892B1 (en) | Apparatus and methods for determining a hierarchical listing of information gaps | |
| US20240070130A1 (en) | Methods And Systems For Identifying And Correcting Anomalies In A Data Environment | |
| US12079291B1 (en) | Apparatus for enhanced outreach and method of use | |
| CN118626701B (en) | Real-time social listening data mining method, device and electronic device | |
| Sharma et al. | Weighted Ensemble LSTM Model with Word Embedding Attention for E-Commerce Product Recommendation | |
| Jeevitha et al. | Using machine learning to identify instances of cyberbullying on social media |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| REG | Reference to a national code |
Ref country code: HK Ref legal event code: DE Ref document number: 40098045 Country of ref document: HK |