CN112667791A - Latent event prediction method, device, equipment and storage medium - Google Patents
Latent event prediction method, device, equipment and storage medium Download PDFInfo
- Publication number
- CN112667791A CN112667791A CN202011539228.XA CN202011539228A CN112667791A CN 112667791 A CN112667791 A CN 112667791A CN 202011539228 A CN202011539228 A CN 202011539228A CN 112667791 A CN112667791 A CN 112667791A
- Authority
- CN
- China
- Prior art keywords
- event
- potential
- potential event
- prediction
- training
- 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
- 238000000034 method Methods 0.000 title claims abstract description 58
- 238000012549 training Methods 0.000 claims abstract description 188
- 238000012216 screening Methods 0.000 claims abstract description 15
- 238000005192 partition Methods 0.000 claims description 102
- 238000006243 chemical reaction Methods 0.000 claims description 32
- 238000012545 processing Methods 0.000 claims description 13
- 238000010606 normalization Methods 0.000 claims description 6
- 230000009466 transformation Effects 0.000 claims description 2
- 238000004590 computer program Methods 0.000 claims 2
- 238000013473 artificial intelligence Methods 0.000 abstract description 2
- 230000008569 process Effects 0.000 description 17
- 238000010586 diagram Methods 0.000 description 7
- 238000012827 research and development Methods 0.000 description 5
- 230000004044 response Effects 0.000 description 5
- 238000012163 sequencing technique Methods 0.000 description 4
- 238000010276 construction Methods 0.000 description 3
- 230000007246 mechanism Effects 0.000 description 3
- 238000004364 calculation method Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000010845 search algorithm Methods 0.000 description 2
- 230000006399 behavior Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 238000013500 data storage Methods 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 230000001788 irregular Effects 0.000 description 1
- 238000002372 labelling Methods 0.000 description 1
- 238000013508 migration Methods 0.000 description 1
- 230000005012 migration Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000002085 persistent effect Effects 0.000 description 1
- 230000000717 retained effect Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 230000001052 transient effect Effects 0.000 description 1
Images
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
- G06F16/33—Querying
- G06F16/332—Query formulation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/20—Natural language analysis
- G06F40/205—Parsing
- G06F40/211—Syntactic parsing, e.g. based on context-free grammar [CFG] or unification grammars
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/30—Semantic analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q30/00—Commerce
- G06Q30/02—Marketing; Price estimation or determination; Fundraising
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Business, Economics & Management (AREA)
- General Physics & Mathematics (AREA)
- Strategic Management (AREA)
- Development Economics (AREA)
- General Engineering & Computer Science (AREA)
- Economics (AREA)
- Computational Linguistics (AREA)
- Marketing (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Game Theory and Decision Science (AREA)
- Human Resources & Organizations (AREA)
- Finance (AREA)
- Entrepreneurship & Innovation (AREA)
- Accounting & Taxation (AREA)
- General Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Health & Medical Sciences (AREA)
- General Business, Economics & Management (AREA)
- Tourism & Hospitality (AREA)
- Quality & Reliability (AREA)
- Data Mining & Analysis (AREA)
- Operations Research (AREA)
- Databases & Information Systems (AREA)
- Mathematical Physics (AREA)
- Machine Translation (AREA)
Abstract
The invention relates to the field of artificial intelligence and discloses a method, a device, equipment and a storage medium for predicting a potential event. The method comprises the following steps: training a pre-training model for predicting potential events for standby by a first data set of multiple fields and adopting a preset segment-by-segment learning paradigm; respectively converting the second data set of the target field into training samples of two formats, and training the training samples in two pre-training models to obtain a second potential event prediction model and a third potential event prediction model for predicting a second potential event corresponding to the second event sentence and a third occurrence probability corresponding to the second potential event sentence; and comparing the second occurrence probability with the third occurrence probability, and screening out a plurality of potential events from the second potential event and the third potential event for pushing based on the comparison result. The invention also relates to a blockchain technique, said data sets being stored in a blockchain. The method and the device realize the prediction of the occurrence of the potential event according to the event sentence in the open scene.
Description
Technical Field
The present invention relates to the field of artificial intelligence, and in particular, to a method, an apparatus, a device, and a storage medium for predicting a potential event.
Background
With the increasing popularity of the existing online services, the increasing number of users and the increasing requirements of enterprises on intelligent robot anthropomorphic, how to handle the response of the open scene becomes an important appeal. Such new scenarios require the machine to "actively" guess some direction that may push a topic, generating a corresponding reply, via the current user's answer. Such application scenarios require the model to satisfy two points: 1) memorizing certain experience and common knowledge; 2) and a prediction result can be generated by combining the newly transmitted samples according to the memorized information. The parameters of the model are enabled to obtain the optimal weight in large-scale label-free data training, and answers are directly generated without classification. That is, the model may provide event response prediction in an open scenario.
However, many application scenarios (e.g. intelligent marketing, intelligent consultation, intelligent debt, complaint response) based on multiple rounds of conversations are often closed, i.e. scene domain specific, conversation target specific, and most conversations can be processed through a simple intention recognition model in combination with a rule template. That is, the existing method cannot predict other things that may happen next after a certain event, so that the existing method has no risk-looking forward capability.
Disclosure of Invention
The invention mainly aims to solve the technical problem that a model can provide event response prediction in an open scene.
The first aspect of the present invention provides a method for predicting a potential event, including:
acquiring a first data set for predicting potential events in multiple fields, adopting a preset segment-by-segment learning paradigm, training the first data set through a preset initial model, predicting a first potential event associated with each first event sentence in the first data set and a first occurrence probability of the first potential event, and stopping until the initial model converges to obtain a pre-training model;
acquiring a second data set of potential event prediction in a target field, and converting the second data set into a first training sample and a second training sample in different model input formats;
fine-tuning the pre-training model by using the first training sample to obtain a first potential event prediction model, and fine-tuning the pre-training model by using the second training sample to obtain a second potential event prediction model;
acquiring a second event sentence to be predicted, inputting the second event sentence into the first potential event prediction model for prediction to obtain a second potential event and a second occurrence probability of the second potential event, and inputting the second event sentence into the second potential event prediction model for prediction to obtain a third potential event and a third occurrence probability of the third potential event;
and comparing the second occurrence probability with the third occurrence probability, and pushing potential events according to a comparison result.
Optionally, in a first implementation manner of the first aspect of the present invention, the acquiring a first data set of predictions of potential events in multiple domains includes:
acquiring original data sets of potential event prediction in multiple fields, and performing abstract definition on the original data sets;
and carrying out normalization processing on the input format of the original data set after the abstract definition to obtain a first data set of potential event prediction in multiple fields.
Optionally, in a second implementation manner of the first aspect of the present invention, the training the first data set by using a preset segment-by-segment learning paradigm and a preset initial model, predicting a first potential event associated with each first event sentence in the first data set and a first occurrence probability of the first potential event, and stopping until the initial model converges, to obtain a pre-trained model, including:
dividing the first data set into a plurality of segments through the preset initial model, and sequentially predicting a first potential event associated with each first event sentence in each segment and a first occurrence probability of the first potential event to obtain a prediction result;
calculating a loss value of a preset initial model based on the prediction result of each segment, and judging whether the loss value is smaller than a preset loss threshold value;
if the loss value is smaller than the preset loss threshold value, determining that the preset initial model is converged to obtain a pre-training model for standby, otherwise, skipping to execute the step of sequentially predicting the first potential event associated with each first event sentence in each segment and the first occurrence probability of the first potential event to obtain a prediction result, and stopping until the loss value is smaller than the preset loss threshold value to obtain the pre-training model.
Optionally, in a third implementation manner of the first aspect of the present invention, the sequentially predicting a first potential event associated with each first event sentence in each segment and a first occurrence probability of the first potential event, and obtaining a prediction result includes:
predicting a first potential event associated with each first event sentence in the current segment and a first occurrence probability of the first potential event to obtain a prediction result of the current segment;
predicting a first potential event associated with each first event sentence in a next segment and a first occurrence probability of the first potential event based on a prediction result of the current segment, and updating the prediction result of the current segment;
and taking the next segment as a current segment, repeatedly executing the step of predicting the first potential event associated with each first event sentence in the next segment and the first occurrence probability of the first potential event until all the segments are predicted, and taking the prediction result of the last segment as a final prediction result.
Optionally, in a fourth implementation manner of the first aspect of the present invention, the converting the second data set into the first training sample and the second training sample of different model input formats includes:
searching a first conversion sample template and a second conversion sample template of the second data set according to a preset model input format;
extracting a plurality of third event sentences in the second data set and a plurality of fourth potential events associated with the third event sentences;
sequentially combining the third event sentences and the associated fourth potential events based on the first conversion sample template to obtain a first training sample, wherein one third event sentence is combined with one fourth potential event in the first training sample;
and sequentially combining each third event sentence in the second data set and each associated fourth potential event based on the second conversion sample template to obtain a second training sample, wherein in the second training sample, one third event sentence is combined with a plurality of fourth potential events.
Optionally, in a fifth implementation manner of the first aspect of the present invention, the sequentially combining each third event sentence in the second data set and each associated fourth potential event based on the second conversion sample template to obtain a second training sample includes:
respectively randomly ordering a plurality of fourth potential events corresponding to each third event sentence in the second data set to obtain a random ordering result, and screening a plurality of second potential events from the random ordering result;
and sequentially combining each third event sentence in the second data set and a fourth potential event obtained by corresponding screening based on the second conversion sample template to obtain a second training sample.
Optionally, in a sixth implementation manner of the first aspect of the present invention, the comparing the second occurrence probability with the third occurrence probability, and performing potential event pushing according to a comparison result includes:
dividing the second occurrence probability into a plurality of first partition probabilities according to occurrence conditions of a second potential event, and dividing the third occurrence probability into a plurality of partition probability sets according to occurrence conditions of a third potential event, wherein the partition probability sets comprise a plurality of second partition probabilities;
sequentially judging whether the first partition probability and each second partition probability under the same occurrence condition are both greater than a preset probability threshold value, and sequentially comparing the first partition probability and each second partition probability under the same occurrence condition;
if the first partition probability and the second partition probabilities are both greater than a preset probability threshold value and the first partition probability is smaller than the minimum second partition probability, selecting a preset number of third potential events to push according to the second partition probability from high to low;
and if the first partition probability and the second partition probabilities are both greater than the probability threshold value, and the first partition probability is greater than the smallest second partition probability, selecting the corresponding second potential event and a third potential event in a second partition probability interval from the first partition probability to the largest second partition probability for pushing.
A second aspect of the present invention provides a potential event prediction apparatus, including:
the system comprises a first training module, a second training module and a third training module, wherein the first training module is used for acquiring a first data set for predicting potential events in multiple fields, adopting a preset segment-by-segment learning paradigm, training the first data set through a preset initial model, predicting a first potential event associated with each first event sentence in the first data set and a first occurrence probability of the first potential event, and stopping until the initial model converges to obtain a pre-training model;
the conversion module is used for acquiring a second data set of potential event prediction in the target field and converting the second data set into a first training sample and a second training sample in different model input formats;
the second training module is used for carrying out fine tuning on the pre-training model by adopting the first training sample to obtain a first potential event prediction model, and carrying out fine tuning on the pre-training model by adopting the second training sample to obtain a second potential event prediction model;
the prediction module is used for acquiring a second event sentence to be predicted, inputting the second event sentence into the first potential event prediction model for prediction to obtain a second potential event and a second occurrence probability of the second potential event, and inputting the second event sentence into the second potential event prediction model for prediction to obtain a third potential event and a third occurrence probability of the third potential event;
and the pushing module is used for comparing the second occurrence probability with the third occurrence probability and pushing potential events according to the comparison result.
Optionally, in a first implementation manner of the second aspect of the present invention, the first training module includes:
the system comprises an abstract processing unit, a data processing unit and a data processing unit, wherein the abstract processing unit is used for acquiring original data sets of potential event prediction in a plurality of fields and performing abstract definition on the original data sets;
and the standardization processing unit is used for carrying out normalization processing on the input format of the original data set after the abstract definition to obtain a first data set for predicting potential events in multiple fields.
Optionally, in a second implementation manner of the second aspect of the present invention, the first training module further includes:
a dividing unit, configured to divide the first data set into multiple segments through the preset initial model, and predict a first potential event associated with each first event sentence in each segment and a first occurrence probability of the first potential event in sequence to obtain a prediction result;
the calculation unit is used for calculating a loss value of a preset initial model based on the prediction result of each segment and judging whether the loss value is smaller than a preset loss threshold value or not;
and a circulating unit, configured to determine that a preset initial model converges if the loss value is smaller than a preset loss threshold, to obtain a pre-trained model for standby, otherwise, skip to perform the step of sequentially predicting a first potential event associated with each first event sentence in each segment and a first occurrence probability of the first potential event, to obtain a prediction result, and stop until the loss value is smaller than the preset loss threshold, to obtain the pre-trained model.
Optionally, in a third implementation manner of the second aspect of the present invention, the dividing unit is further configured to:
predicting a first potential event associated with each first event sentence in the current segment and a first occurrence probability of the first potential event to obtain a prediction result of the current segment;
predicting a first potential event associated with each first event sentence in a next segment and a first occurrence probability of the first potential event based on a prediction result of the current segment, and updating the prediction result of the current segment;
and taking the next segment as a current segment, repeatedly executing the step of predicting the first potential event associated with each first event sentence in the next segment and the first occurrence probability of the first potential event until all the segments are predicted, and taking the prediction result of the last segment as a final prediction result.
Optionally, in a fourth implementation manner of the second aspect of the present invention, the conversion module includes:
the searching unit is used for searching a first conversion sample template and a second conversion sample template of the second data set according to a preset model input format;
an extracting unit, configured to extract a plurality of third event sentences in the second data set and a plurality of fourth potential events associated with the third event sentences;
a first combining unit, configured to sequentially combine the third event sentences and the associated fourth potential events based on the first conversion sample template to obtain a first training sample, where in the first training sample, one third event sentence is combined with one fourth potential event;
and a second combining unit, configured to sequentially combine each third event sentence in the second data set with each associated fourth potential event based on the second conversion sample template to obtain a second training sample, where in the second training sample, one third event sentence is combined with multiple fourth potential events.
Optionally, in a fifth implementation manner of the second aspect of the present invention, the second combining unit is further configured to:
respectively randomly ordering a plurality of fourth potential events corresponding to each third event sentence in the second data set to obtain a random ordering result, and screening a plurality of second potential events from the random ordering result;
and sequentially combining each third event sentence in the second data set and a fourth potential event obtained by corresponding screening based on the second conversion sample template to obtain a second training sample.
Optionally, in a sixth implementation manner of the second aspect of the present invention, the pushing module includes:
the dividing unit is used for dividing the second occurrence probability into a plurality of first partition probabilities according to the occurrence condition of a second potential event and dividing the third occurrence probability into a plurality of partition probability sets according to the occurrence condition of a third potential event, wherein the partition probability sets comprise a plurality of second partition probabilities;
the judging unit is used for sequentially judging whether the first partition probability and the second partition probabilities under the same occurrence condition are both larger than a preset probability threshold value or not and sequentially comparing the first partition probability and the second partition probabilities under the same occurrence condition;
the pushing unit is used for selecting a preset number of third potential events to push according to the second partition probability from high to low if the first partition probability and the second partition probabilities are both greater than a preset probability threshold and the first partition probability is smaller than the minimum second partition probability; and if the first partition probability and the second partition probabilities are both greater than the probability threshold value, and the first partition probability is greater than the smallest second partition probability, selecting the corresponding second potential event and a third potential event in a second partition probability interval from the first partition probability to the largest second partition probability for pushing.
A third aspect of the present invention provides a potential event prediction apparatus comprising: a memory and at least one processor, the memory having instructions stored therein; the at least one processor invokes the instructions in the memory to cause the potential event prediction device to perform the potential event prediction method described above.
A fourth aspect of the present invention provides a computer-readable storage medium having stored therein instructions, which, when run on a computer, cause the computer to execute the above-described potential event prediction method.
In the technical scheme provided by the invention, in the embodiment of the invention, a pre-training model is trained for standby through a first data set of a plurality of fields; then, in the research and development process, a first potential event prediction model and a second potential event prediction model are trained through a second data set of the target field; secondly, in the application process, based on a third event sentence to be predicted and occurrence conditions, predicting second and third potential events corresponding to the third event sentence and corresponding second and third occurrence probabilities by adopting a first and second potential event prediction models; and finally, screening a plurality of potential events from the second and third potential events according to the second and third occurrence probabilities, and pushing the potential events. According to the invention, the priori knowledge is embedded into the model in a mode of combining pre-training with migration without introducing a knowledge base, various exceptional situations can be more flexibly processed, the generalization capability of the model is strong, the occurrence of potential events is predicted according to event sentences in an open scene, the tedious work that a service especially makes answers aiming at irregular problems of users is avoided, and the labor cost is saved.
Drawings
FIG. 1 is a diagram of a first embodiment of a method for predicting potential events according to an embodiment of the present invention;
FIG. 2 is a diagram of a second embodiment of a method for predicting potential events according to an embodiment of the present invention;
FIG. 3 is a diagram of a third embodiment of a method for predicting a potential event according to an embodiment of the present invention;
FIG. 4 is a schematic diagram of an embodiment of a potential event prediction device in an embodiment of the present invention;
FIG. 5 is a schematic diagram of another embodiment of a potential event prediction device in an embodiment of the present invention;
fig. 6 is a schematic diagram of an embodiment of a potential event prediction device in an embodiment of the present invention.
Detailed Description
The embodiment of the invention provides a method, a device, equipment and a storage medium for predicting a potential event, wherein a pre-training model for predicting the potential event is trained for standby by a first data set in multiple fields and a preset segment-by-segment learning paradigm; respectively converting the second data set of the target field into training samples of two formats, and training the training samples in two pre-training models to obtain a first potential event prediction model and a second potential event prediction model for predicting a third potential event corresponding to a third event sentence to be predicted and a corresponding first occurrence probability and a corresponding second occurrence probability; and comparing the first occurrence probability with the second occurrence probability, and screening out a plurality of potential events from the third potential event and the fourth potential event for pushing based on the comparison result. The method and the device realize the prediction of the occurrence of the potential event according to the event sentence in the open scene.
The terms "first," "second," "third," "fourth," and the like in the description and in the claims, as well as in the drawings, if any, are used for distinguishing between similar elements and not necessarily for describing a particular sequential or chronological order. It will be appreciated that the data so used may be interchanged under appropriate circumstances such that the embodiments described herein may be practiced otherwise than as specifically illustrated or described herein. Furthermore, the terms "comprises," "comprising," or "having," and any variations thereof, are intended to cover non-exclusive inclusions, such that a process, method, system, article, or apparatus that comprises a list of steps or elements is not necessarily limited to those steps or elements expressly listed, but may include other steps or elements not expressly listed or inherent to such process, method, article, or apparatus.
For ease of understanding, a detailed flow of an embodiment of the present invention is described below, and referring to fig. 1, a first embodiment of a method for predicting a latent event according to an embodiment of the present invention includes:
101. acquiring a first data set for predicting potential events in multiple fields, adopting a preset segment-by-segment learning paradigm, training the first data set through a preset initial model, predicting a first potential event associated with each first event sentence in the first data set and a first occurrence probability of the first potential event, and stopping until the initial model converges to obtain a pre-training model;
it is to be understood that the execution subject of the present invention may be a potential event prediction device, and may also be a terminal or a server, which is not limited herein. The embodiment of the present invention is described by taking a server as an execution subject. It is emphasized that to further ensure the privacy and security of each data set, each data set may also be stored in a node of a blockchain.
In this embodiment, the first data set includes various types of documents in different fields, including news and standard documents such as economy, finance, sports, science and technology, entertainment, animation, etc., and other language documents besides the Chinese may be used, including english, french, german, japanese, korean, etc., according to the scale of the first data set; the storage format of each piece of data in the first data set is as follows: one event sentence + one potential event condition + a plurality of potential events. The term "event sentence" refers to a descriptive sentence of a currently occurring event, the term "potential event condition" refers to a condition of a potential event, and the term "potential event" refers to an expected event that will occur under a specific potential event condition. For example, the "event sentence" presents a gift to the small white for the small king, "the" potential event condition "is how the small white will do, and the" potential event "is the" small white thanks to the small king ".
The method comprises the steps of firstly training a pre-training model for predicting the potential events by adopting a first data set relating to a plurality of fields, and then carrying out fine adjustment on the data set of the target field when the pre-training model is applied to a specific target field, wherein the input of the application model after the fine adjustment of the pre-training model is an event sentence, and the output of the application model is the potential event associated with the event sentence and a first occurrence probability.
102. Acquiring a second data set of potential event prediction in a target field, and converting the second data set into a first training sample and a second training sample in different model input formats;
in this embodiment, when performing the prediction of the potential event in the specific target domain, the pre-training model may be fine-tuned by the second data set related to the target domain. And converting the second data set into training samples with fixed input formats according to the preset model input format, and training two potential event prediction models by adopting two training samples, so that the accuracy and the generation efficiency of the final input prediction result can be improved. When the method is applied to a certain target field, only a small number of second data sets of the target field need to be collected, and then supervised learning is carried out, so that the labeling workload of training samples is reduced, and the model training efficiency is improved.
103. Fine-tuning the pre-training model by using the first training sample to obtain a first potential event prediction model, and fine-tuning the pre-training model by using the second training sample to obtain a second potential event prediction model;
in this embodiment, after the first and second training samples are input into the pre-training model, the fine tuning process is the same as the training process, and only the input training samples are different and the input results are different. The content of the first training sample is an event sentence + a potential event occurrence condition + a potential event, and the potential event and the corresponding occurrence probability are input after the pre-training model training; the content of the second training sample is an event sentence + a potential event occurrence condition + a plurality of potential events, and after the pre-training model training, a plurality of potential events and the occurrence probability corresponding to each potential event are input; the latter is more concerned with the overall semantic information than the former, and the final output is mainly based on the input potential events of the latter.
In addition, in the model training process, the key is the decoding strategy of the decoder. Here the training of both pre-trained models is based on the Greedy algorithm (Greedy algorithm) and the beam search algorithm (BeamSearch algorithm). Selecting characters which are most consistent (namely potential events corresponding to the maximum probability value predicted by the model) from the prior knowledge attached to the pre-training model according to the output text by the greedy algorithm each time, and connecting the characters to the input text; and the beam search algorithm further adds a deduplication mechanism (sliding window calculating n-gram, if the n-gram value identical to the previous text exists, a penalty value is applied to the repeated value) and a filtering mechanism (kernel method: a threshold line of the BeamSearch value is defined, and subsequent calculation can be carried out only when the score exceeds the result of the line).
104. Acquiring a second event sentence to be predicted, inputting the second event sentence into the first potential event prediction model for prediction to obtain a second potential event and a second occurrence probability of the second potential event, and inputting the second event sentence into the second potential event prediction model for prediction to obtain a third potential event and a third occurrence probability of the third potential event;
in the embodiment, in practical application, according to the second event sentence to be predicted, the occurrence condition of the potential event is extracted and the potential event is predicted, for example, in intelligent marketing, commodities are recommended according to the retrieval history of the user; predicting answers according to questions input by a user in the intelligent consultation; predicting repayment intention, repayment capacity, repayment time and the like of the user according to conversation voice content of the user in the intelligent debt urging and answering; in the complaint response, the user, the possible non-goodwill behaviors of the complaint object, the predicted solution, and the like are predicted according to the complaint content of the user.
105. And comparing the second occurrence probability with the third occurrence probability, and pushing potential events according to a comparison result.
In this embodiment, the first and second potential event prediction models respectively output potential events and occurrence probabilities corresponding to an event sentence to be predicted, theoretically, there are two different sets of prediction results, that is, different potential events and occurrence probabilities, where the second potential event includes one potential event, the second occurrence probability includes the occurrence probability of one potential event, the third potential event includes multiple potential events, and the third occurrence probability includes the occurrence probabilities of multiple potential events, and the second occurrence probability and the third occurrence probability are compared, and according to the magnitude of the two probabilities, a set of preset number of potential events is screened out as the most likely event to occur after the current event to be predicted.
In the embodiment of the invention, a pre-training model is trained for standby through a first data set of a plurality of fields; then, in the research and development process, a first potential event prediction model and a second potential event prediction model are trained through a second data set of the target field; secondly, in the application process, based on a third event sentence to be predicted and occurrence conditions, predicting second and third potential events corresponding to the third event sentence and corresponding second and third occurrence probabilities by adopting a first and second potential event prediction models; and finally, screening a plurality of potential events from the second and third potential events according to the second and third occurrence probabilities, and pushing the potential events.
Referring to fig. 2, a second embodiment of the method for predicting a latent event according to the present invention includes:
201. acquiring original data sets of potential event prediction in multiple fields, and performing abstract definition on the original data sets;
in this embodiment, in order to ensure that the sample forms during model application and training are consistent, it is necessary to abstractly define the input original data set, for example, specific names in the input sentence may be replaced with abstract names such as PersonX and PersonY. Finally, when the result is generated, the abstract person names such as PersonX, PersonY and the like are changed back to the real person name.
202. Carrying out normalization processing on an input format on the original data set after abstract definition to obtain a first data set for predicting potential events in multiple fields;
in addition, the segment-by-segment learning paradigm refers to dividing the text segments of the complete semantics existing among the event sentences, the potential conditions and/or the potential events in the first data set, and predicting the next text segment according to the currently known text segments. .
In this embodiment, a pre-training model common to each field is also constructed in advance through unsupervised learning, and the pre-training model is obtained by performing fine tuning on an open-source large-scale unsupervised generation model by using the first data set. The method adopts a self-built and self-training model MTSN-base (multi-port Time-Sensitive network), and an auto-regression encoder-decoder structure (auto-regression encoder-decoder frame) formed based on a transform is trained by adopting a segment-by-segment learning paradigm.
203. Dividing the first data set into a plurality of segments through the preset initial model, and sequentially predicting a first potential event associated with each first event sentence in each segment and a first occurrence probability of the first potential event to obtain a prediction result;
in this embodiment, the preset initial model may predict the first potential event associated with each first event sentence and the first occurrence probability of each first potential event in each segment through the following steps:
(1) predicting a first potential event associated with each first event sentence in the current segment and a first occurrence probability of the first potential event to obtain a prediction result of the current segment;
(2) predicting a first potential event associated with each first event sentence in a next segment and a first occurrence probability of the first potential event based on a prediction result of the current segment, and updating the prediction result of the current segment;
(3) and taking the next segment as a current segment, repeatedly performing the first potential event prediction and the first occurrence probability prediction until all the segments are predicted, and taking the prediction result of the last segment as a final prediction result.
In the embodiment, the priori knowledge of the pre-training model is constructed as the prediction result through unsupervised learning of each segment in the first data set, and the potential occurrence event of the input event sentence can be generated through the priori knowledge in the later period when application training or actual use in a certain field is specifically carried out.
204. Calculating a loss value of a preset initial model based on the prediction result of each segment, and judging whether the loss value is smaller than a preset loss threshold value;
205. if the loss value is smaller than the preset loss threshold value, determining that a preset initial model is converged to obtain a pre-training model for standby, otherwise, skipping to execute the step of sequentially predicting a first potential event associated with each first event sentence in each segment and the first occurrence probability of the first potential event to obtain a prediction result, and stopping until the loss value is smaller than the preset loss threshold value to obtain the pre-training model;
in this embodiment, the first data set is divided into a plurality of segments to perform segment-by-segment learning, and unlike the conventional word-by-word learning method, the segment-by-segment learning paradigm allows semantic information in the training sample to be retained, so that the semantic information does not depend too much on a word at a previous predicted position, and the difference between the semantic information and a real writing habit is small.
206. Acquiring a second data set of potential event prediction in a target field, and converting the second data set into a first training sample and a second training sample in different model input formats;
207. fine-tuning the pre-training model by using the first training sample to obtain a first potential event prediction model, and fine-tuning the pre-training model by using the second training sample to obtain a second potential event prediction model;
208. acquiring a second event sentence to be predicted, inputting the second event sentence into the first potential event prediction model for prediction to obtain a second potential event and a second occurrence probability of the second potential event, and inputting the second event sentence into the second potential event prediction model for prediction to obtain a third potential event and a third occurrence probability of the third potential event;
209. and comparing the second occurrence probability with the third occurrence probability, and pushing potential events according to a comparison result.
In the embodiment of the invention, the pre-training model for predicting the potential event is trained by adopting the first data sets in multiple fields in a research and development stage in a segment-by-segment learning paradigm, so that the potential event prediction model obtained by subsequent fine tuning not only considers the single character of an event sentence, but also considers the actual artificial habit, the predicted potential event is closer to the actual artificial condition when the event sentence is input, and the prediction accuracy of the potential event in different scenes is improved.
Referring to fig. 3, a fourth embodiment of the method for predicting a latent event according to the present invention includes:
301. acquiring a first data set for predicting potential events in multiple fields, adopting a preset segment-by-segment learning paradigm, training the first data set through a preset initial model, predicting a first potential event associated with each first event sentence in the first data set and a first occurrence probability of the first potential event, and stopping until the initial model converges to obtain a pre-training model;
302. searching a first conversion sample template and a second conversion sample template of the second data set according to a preset model input format;
in this embodiment, a specific model input format is as follows:
A. the model input format of the first training sample is: [ CLS ] event sentence [ input Condition ] [ SEP ] latent event [ END ]. Where [ CLS ] is a prediction prompt, [ SEP ] indicates a division input condition and a potential event, and [ END ] indicates the END of generation.
B. The model input format of the second training sample is: [ CLS]Event sentence [ input conditions ]][SEP]Latent event A1[ separator ]]Latent event A2[ separator ]]… … latent event An[END]。
303. Extracting a plurality of third event sentences in the second data set and a plurality of fourth potential events associated with the third event sentences;
304. sequentially combining the third event sentences and the associated fourth potential events based on the first conversion sample template to obtain a first training sample, wherein one third event sentence is combined with one fourth potential event in the first training sample;
305. on the basis of the second conversion sample template, sequentially combining each third event sentence in the second data set with each associated fourth potential event to obtain a second training sample, wherein in the second training sample, one third event sentence is combined with a plurality of fourth potential events;
in this embodiment, the fixed combination format of each piece of training data is determined by the model input format of the first training sample and the model input format of the second training sample, and each piece of training data of the first training sample is: an event sentence + a potential event occurrence condition + a potential event, and the training data of the second training sample are: an event sentence + a potential event occurrence condition + a plurality of potential events; wherein the specific occurring potential event corresponds to the potential event occurrence condition, and the specific transformation process of the second training sample is as follows:
(1) and respectively randomly sequencing a plurality of fourth potential events corresponding to each third event sentence in the second data set, and randomly screening a plurality of second potential events from a random sequencing result in sequence.
(2) And sequentially combining each third event sentence in the second data set and a fourth potential event obtained by corresponding screening based on the second conversion sample template to obtain a second training sample.
The potential events are randomly ordered first, and part (preset number) of the potential events are screened for subsequent model training, so that the generalization capability of the pre-trained model can be prompted when a second training sample is trained.
306. Fine-tuning the pre-training model by using the first training sample to obtain a first potential event prediction model, and fine-tuning the pre-training model by using the second training sample to obtain a second potential event prediction model;
307. acquiring a second event sentence to be predicted, inputting the second event sentence into the first potential event prediction model for prediction to obtain a second potential event and a second occurrence probability of the second potential event, and inputting the second event sentence into the second potential event prediction model for prediction to obtain a third potential event and a third occurrence probability of the third potential event;
308. and comparing the second occurrence probability with the third occurrence probability, and pushing potential events according to a comparison result.
In this embodiment, a plurality of potential events may be specifically screened from the second and third potential events by the following rules to be pushed:
(1) dividing the second occurrence probability into a plurality of first partition probabilities according to occurrence conditions of a second potential event, and dividing the third occurrence probability into a plurality of partition probability sets according to occurrence conditions of a third potential event, wherein the partition probability sets comprise a plurality of second partition probabilities;
(2) sequentially judging whether the first partition probability and each second partition probability under the same occurrence condition are both greater than a preset probability threshold value, and sequentially comparing the first partition probability and each second partition probability under the same occurrence condition;
(3) if the first partition probability and the second partition probabilities are both greater than a preset probability threshold value and the first partition probability is smaller than the minimum second partition probability, selecting a preset number of third potential events to push according to the second partition probability from high to low;
(4) and if the first partition probability and the second partition probabilities are both greater than the probability threshold value, and the first partition probability is greater than the smallest second partition probability, selecting the corresponding second potential event and a third potential event in a second partition probability interval from the first partition probability to the largest second partition probability for pushing.
And if the first partition probability and the second partition probability are both smaller than a preset probability threshold, determining that no potential event meeting the conditions exists, and directly returning to None.
Preferably, the preset probability threshold may be selected to be 0.7 and the preset number may be selected to be 3.
In the embodiment of the invention, the construction process of a first training sample and a second training sample is introduced in detail for training two types of potential event prediction models, a second potential event and a third potential event which are related to a second phase and corresponding second and third occurrence probabilities are predicted according to an input second event sentence through the first and second potential event prediction models, and the potential event with the highest occurrence probability is screened by combining the second and third occurrence probabilities, so that the accuracy of potential event prediction is improved.
With reference to fig. 4, the method for predicting a potential event in the embodiment of the present invention is described above, and a device for predicting a potential event in the embodiment of the present invention is described below, where an embodiment of the device for predicting a potential event in the embodiment of the present invention includes:
a first training module 401, configured to obtain a first data set for predicting potential events in multiple fields, train the first data set by using a preset segment-by-segment learning paradigm, predict a first potential event associated with each first event sentence in the first data set and a first occurrence probability of the first potential event, and stop until the initial model converges to obtain a pre-training model;
a conversion module 402, configured to obtain a second data set for predicting a potential event in a target field, and convert the second data set into a first training sample and a second training sample in different model input formats;
a second training module 403, configured to perform fine tuning on the pre-training model by using the first training sample to obtain a first potential event prediction model, and perform fine tuning on the pre-training model by using the second training sample to obtain a second potential event prediction model;
a predicting module 404, configured to obtain a second event sentence to be predicted, input the second event sentence into the first potential event prediction model for prediction, to obtain a second potential event and a second occurrence probability of the second potential event, and input the second event sentence into the second potential event prediction model for prediction, to obtain a third potential event and a third occurrence probability of the third potential event;
and a pushing module 405, configured to compare the second occurrence probability with the third occurrence probability, and push a potential event according to a comparison result.
In the embodiment of the invention, a pre-training model is trained for standby through a first data set of a plurality of fields; then, in the research and development process, a first potential event prediction model and a second potential event prediction model are trained through a second data set of the target field; secondly, in the application process, based on a third event sentence to be predicted and occurrence conditions, predicting second and third potential events corresponding to the third event sentence and corresponding second and third occurrence probabilities by adopting a first and second potential event prediction models; and finally, screening a plurality of potential events from the second and third potential events according to the second and third occurrence probabilities, and pushing the potential events.
Referring to fig. 5, another embodiment of the latent event predicting apparatus according to the embodiment of the present invention includes:
a first training module 401, configured to obtain a first data set for predicting potential events in multiple fields, train the first data set by using a preset segment-by-segment learning paradigm, predict a first potential event associated with each first event sentence in the first data set and a first occurrence probability of the first potential event, and stop until the initial model converges to obtain a pre-training model;
a conversion module 402, configured to obtain a second data set for predicting a potential event in a target field, and convert the second data set into a first training sample and a second training sample in different model input formats;
a second training module 403, configured to perform fine tuning on the pre-training model by using the first training sample to obtain a first potential event prediction model, and perform fine tuning on the pre-training model by using the second training sample to obtain a second potential event prediction model;
a predicting module 404, configured to obtain a second event sentence to be predicted, input the second event sentence into the first potential event prediction model for prediction, to obtain a second potential event and a second occurrence probability of the second potential event, and input the second event sentence into the second potential event prediction model for prediction, to obtain a third potential event and a third occurrence probability of the third potential event;
and a pushing module 405, configured to compare the second occurrence probability with the third occurrence probability, and push a potential event according to a comparison result.
Specifically, the first training module includes:
the abstract processing unit 4011 is configured to obtain an original data set for predicting potential events in multiple fields, and perform abstract definition on the original data set;
the normalization processing unit 4012 is configured to perform normalization processing on the input format of the raw data set after the abstract definition, so as to obtain a first data set for predicting potential events in multiple fields.
Specifically, the first training module further includes:
a dividing unit 4013, configured to divide the first data set into a plurality of segments according to the preset initial model, and predict a first potential event associated with each first event sentence in each segment and a first occurrence probability of the first potential event in sequence to obtain a prediction result;
the calculating unit 4014 is configured to calculate a loss value of a preset initial model based on the prediction result of each segment, and determine whether the loss value is smaller than a preset loss threshold;
and the circulating unit 4015 is configured to determine that a preset initial model converges if the loss value is smaller than a preset loss threshold, to obtain a pre-trained model for standby, otherwise, skip to perform the step of sequentially predicting a first potential event associated with each first event sentence in each segment and a first occurrence probability of the first potential event, to obtain a prediction result, and stop until the loss value is smaller than the preset loss threshold, to obtain the pre-trained model.
Specifically, the dividing unit 4013 is further configured to:
predicting a first potential event associated with each first event sentence in the current segment and a first occurrence probability of the first potential event to obtain a prediction result of the current segment;
predicting a first potential event associated with each first event sentence in a next segment and a first occurrence probability of the first potential event based on a prediction result of the current segment, and updating the prediction result of the current segment;
and taking the next segment as a current segment, repeatedly performing the first potential event prediction and the first occurrence probability prediction until all the segments are predicted, and taking the prediction result of the last segment as a final prediction result.
Specifically, the conversion module 402 includes:
the searching unit 4021 is configured to search the first conversion sample template and the second conversion sample template of the second data set according to a preset model input format;
an extracting unit 4022, configured to extract a plurality of third event sentences in the second data set and a plurality of fourth potential events associated with the third event sentences;
a first combining unit 4023, configured to sequentially combine the third event sentences and the associated fourth potential events based on the first conversion sample template to obtain a first training sample, where in the first training sample, one third event sentence is combined with one fourth potential event;
a second combining unit 4024, configured to sequentially combine each third event sentence in the second data set with each associated fourth potential event based on the second conversion sample template to obtain a second training sample, where in the second training sample, one third event sentence is combined with multiple fourth potential events.
Specifically, the second combination unit 4024 is further configured to:
respectively randomly sequencing a plurality of fourth potential events corresponding to each third event sentence in the second data set, and randomly screening a plurality of second potential events from a random sequencing result in sequence;
and sequentially combining each third event sentence in the second data set and a fourth potential event obtained by corresponding screening based on the second conversion sample template to obtain a second training sample.
Specifically, the pushing module 405 includes:
a dividing unit 4051, configured to divide the second occurrence probability into a plurality of first partition probabilities according to an occurrence condition of a second potential event, and divide the third occurrence probability into a plurality of partition probability sets according to an occurrence condition of a third potential event, where each partition probability set includes a plurality of second partition probabilities;
a judging unit 4052, configured to sequentially judge whether the first partition probability and the second partition probabilities under the same occurrence condition are both greater than a preset probability threshold, and sequentially compare the first partition probability and the second partition probabilities under the same occurrence condition;
a pushing unit 4053, configured to select a preset number of third potential events to be pushed according to a second partition probability from high to low if the first partition probability and the second partition probabilities are both greater than a preset probability threshold and the first partition probability is smaller than the minimum second partition probability; and if the first partition probability and the second partition probabilities are both greater than the probability threshold value, and the first partition probability is greater than the smallest second partition probability, selecting the corresponding second potential event and a third potential event in a second partition probability interval from the first partition probability to the largest second partition probability for pushing.
In the embodiment of the invention, a pre-training model for predicting a potential event is trained by adopting a first data set in multiple fields in a research and development stage in a segment-by-segment learning paradigm, so that a potential event prediction model obtained by subsequent fine tuning not only considers the single character of an event sentence, but also considers the actual artificial habit, and the predicted potential event is closer to the actual artificial condition when the event sentence is input; the construction process of the first training sample and the second training sample is also introduced in detail at the same time, so that the construction process is used for training two types of potential event prediction models, the second and third potential events related to the second phase and the corresponding second and third occurrence probabilities are predicted according to the input second event sentence through the first and second potential event prediction models, the potential event with the highest occurrence probability is screened by combining the second and third occurrence probabilities, and the accuracy of potential event prediction is improved.
Fig. 4 and 5 describe the potential event prediction apparatus in the embodiment of the present invention in detail from the perspective of the modular functional entity, and the potential event prediction device in the embodiment of the present invention is described in detail from the perspective of hardware processing.
Fig. 6 is a schematic structural diagram of a potential event prediction device according to an embodiment of the present invention, where the potential event prediction device 600 may generate relatively large differences due to different configurations or performances, and may include one or more processors (CPUs) 610 (e.g., one or more processors) and a memory 620, and one or more storage media 630 (e.g., one or more mass storage devices) for storing applications 633 or data 632. Memory 620 and storage medium 630 may be, among other things, transient or persistent storage. The program stored in the storage medium 630 may include one or more modules (not shown), each of which may include a sequence of instructions operating on the potential event prediction device 600. Still further, processor 610 may be configured to communicate with storage medium 630 to execute a series of instruction operations in storage medium 630 on potential event prediction device 600.
The potential event prediction device 600 may also include one or more power supplies 640, one or more wired or wireless network interfaces 650, one or more input-output interfaces 660, and/or one or more operating systems 631, such as Windows Server, Mac OS X, Unix, Linux, FreeBSD, and the like. Those skilled in the art will appreciate that the potential event prediction device configuration shown in fig. 6 does not constitute a limitation of the potential event prediction device and may include more or fewer components than shown, or some components may be combined, or a different arrangement of components.
The present invention also provides a potential event prediction device, which includes a memory and a processor, wherein the memory stores computer readable instructions, and the computer readable instructions, when executed by the processor, cause the processor to execute the steps of the potential event prediction method in the above embodiments.
The present invention also provides a computer readable storage medium, which may be a non-volatile computer readable storage medium, and which may also be a volatile computer readable storage medium, having stored therein instructions, which, when executed on a computer, cause the computer to perform the steps of the potential event prediction method.
It is clear to those skilled in the art that, for convenience and brevity of description, the specific working processes of the above-described systems, apparatuses and units may refer to the corresponding processes in the foregoing method embodiments, and are not described herein again.
The block chain is a novel application mode of computer technologies such as distributed data storage, point-to-point transmission, a consensus mechanism, an encryption algorithm and the like. A block chain (Blockchain), which is essentially a decentralized database, is a series of data blocks associated by using a cryptographic method, and each data block contains information of a batch of network transactions, so as to verify the validity (anti-counterfeiting) of the information and generate a next block. The blockchain may include a blockchain underlying platform, a platform product service layer, an application service layer, and the like.
The above-mentioned embodiments are only used for illustrating the technical solutions of the present invention, and not for limiting the same; although the present invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; and such modifications or substitutions do not depart from the spirit and scope of the corresponding technical solutions of the embodiments of the present invention.
Claims (10)
1. A potential event prediction method, comprising:
acquiring a first data set for predicting potential events in multiple fields, adopting a preset segment-by-segment learning paradigm, training the first data set through a preset initial model, predicting a first potential event associated with each first event sentence in the first data set and a first occurrence probability of the first potential event, and stopping until the initial model converges to obtain a pre-training model;
acquiring a second data set of potential event prediction in a target field, and converting the second data set into a first training sample and a second training sample in different model input formats;
fine-tuning the pre-training model by using the first training sample to obtain a first potential event prediction model, and fine-tuning the pre-training model by using the second training sample to obtain a second potential event prediction model;
acquiring a second event sentence to be predicted, inputting the second event sentence into the first potential event prediction model for prediction to obtain a second potential event and a second occurrence probability of the second potential event, and inputting the second event sentence into the second potential event prediction model for prediction to obtain a third potential event and a third occurrence probability of the third potential event;
and comparing the second occurrence probability with the third occurrence probability, and pushing potential events according to a comparison result.
2. The method of claim 1, wherein obtaining the first data set of potential event predictions for a plurality of domains comprises:
acquiring original data sets of potential event prediction in multiple fields, and performing abstract definition on the original data sets;
and carrying out normalization processing on the input format of the original data set after the abstract definition to obtain a first data set of potential event prediction in multiple fields.
3. The method according to claim 1, wherein the training of the first data set through a preset initial model using a preset segment-by-segment learning paradigm to predict a first potential event associated with each first event sentence in the first data set and a first occurrence probability of the first potential event is stopped until the initial model converges, and obtaining a pre-trained model comprises:
dividing the first data set into a plurality of segments through the preset initial model, and sequentially predicting a first potential event associated with each first event sentence in each segment and a first occurrence probability of the first potential event to obtain a prediction result;
calculating a loss value of a preset initial model based on the prediction result of each segment, and judging whether the loss value is smaller than a preset loss threshold value;
if the loss value is smaller than the preset loss threshold value, determining that the preset initial model is converged to obtain a pre-training model for standby, otherwise, skipping to execute the step of sequentially predicting the first potential event associated with each first event sentence in each segment and the first occurrence probability of the first potential event to obtain a prediction result, and stopping until the loss value is smaller than the preset loss threshold value to obtain the pre-training model.
4. The method according to claim 3, wherein the predicting the first potential event associated with each first event sentence in each segment and the first occurrence probability of the first potential event in sequence to obtain the predicted result comprises:
predicting a first potential event associated with each first event sentence in the current segment and a first occurrence probability of the first potential event to obtain a prediction result of the current segment;
predicting a first potential event associated with each first event sentence in a next segment and a first occurrence probability of the first potential event based on a prediction result of the current segment, and updating the prediction result of the current segment;
and taking the next segment as a current segment, repeatedly executing the step of predicting the first potential event associated with each first event sentence in the next segment and the first occurrence probability of the first potential event until all the segments are predicted, and taking the prediction result of the last segment as a final prediction result.
5. The method of potential event prediction according to claim 1, wherein the converting the second data set into first and second training samples of different model input formats comprises:
searching a first conversion sample template and a second conversion sample template of the second data set according to a preset model input format;
extracting a plurality of third event sentences in the second data set and a plurality of fourth potential events associated with the third event sentences;
sequentially combining the third event sentences and the associated fourth potential events based on the first conversion sample template to obtain a first training sample, wherein one third event sentence is combined with one fourth potential event in the first training sample;
and sequentially combining each third event sentence in the second data set and each associated fourth potential event based on the second conversion sample template to obtain a second training sample, wherein in the second training sample, one third event sentence is combined with a plurality of fourth potential events.
6. The method of predicting potential events according to claim 5, wherein the sequentially combining each third event sentence in the second data set with each associated fourth potential event based on the second transformation sample template to obtain a second training sample comprises:
respectively randomly ordering a plurality of fourth potential events corresponding to each third event sentence in the second data set to obtain a random ordering result, and screening a plurality of second potential events from the random ordering result;
and sequentially combining each third event sentence in the second data set and a fourth potential event obtained by corresponding screening based on the second conversion sample template to obtain a second training sample.
7. The method for predicting potential events according to any one of claims 1 to 6, wherein the comparing the second occurrence probability with the third occurrence probability and performing potential event pushing according to the comparison result comprises:
dividing the second occurrence probability into a plurality of first partition probabilities according to occurrence conditions of a second potential event, and dividing the third occurrence probability into a plurality of partition probability sets according to occurrence conditions of a third potential event, wherein the partition probability sets comprise a plurality of second partition probabilities;
sequentially judging whether the first partition probability and each second partition probability under the same occurrence condition are both greater than a preset probability threshold value, and sequentially comparing the first partition probability and each second partition probability under the same occurrence condition;
if the first partition probability and the second partition probabilities are both greater than a preset probability threshold value and the first partition probability is smaller than the minimum second partition probability, selecting a preset number of third potential events to push according to the second partition probability from high to low;
and if the first partition probability and the second partition probabilities are both greater than the probability threshold value, and the first partition probability is greater than the smallest second partition probability, selecting the corresponding second potential event and a third potential event in a second partition probability interval from the first partition probability to the largest second partition probability for pushing.
8. A potential event prediction apparatus, comprising:
the system comprises a first training module, a second training module and a third training module, wherein the first training module is used for acquiring a first data set for predicting potential events in multiple fields, adopting a preset segment-by-segment learning paradigm, training the first data set through a preset initial model, predicting a first potential event associated with each first event sentence in the first data set and a first occurrence probability of the first potential event, and stopping until the initial model converges to obtain a pre-training model;
the conversion module is used for acquiring a second data set of potential event prediction in the target field and converting the second data set into a first training sample and a second training sample in different model input formats;
the second training module is used for carrying out fine tuning on the pre-training model by adopting the first training sample to obtain a first potential event prediction model, and carrying out fine tuning on the pre-training model by adopting the second training sample to obtain a second potential event prediction model;
the prediction module is used for acquiring a second event sentence to be predicted, inputting the second event sentence into the first potential event prediction model for prediction to obtain a second potential event and a second occurrence probability of the second potential event, and inputting the second event sentence into the second potential event prediction model for prediction to obtain a third potential event and a third occurrence probability of the third potential event;
and the pushing module is used for comparing the second occurrence probability with the third occurrence probability and pushing potential events according to the comparison result.
9. A potential event prediction device, characterized in that the potential event prediction device comprises: a memory and at least one processor, the memory having instructions stored therein;
the at least one processor invokes the instructions in the memory to cause the potential event prediction device to perform the potential event prediction method of any of claims 1-7.
10. A computer-readable storage medium, having stored thereon a computer program, wherein the computer program, when executed by a processor, implements a method of potential event prediction according to any of claims 1-7.
Priority Applications (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011539228.XA CN112667791A (en) | 2020-12-23 | 2020-12-23 | Latent event prediction method, device, equipment and storage medium |
| PCT/CN2021/126112 WO2022134834A1 (en) | 2020-12-23 | 2021-10-25 | Potential event predicting method, apparatus and device, and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011539228.XA CN112667791A (en) | 2020-12-23 | 2020-12-23 | Latent event prediction method, device, equipment and storage medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN112667791A true CN112667791A (en) | 2021-04-16 |
Family
ID=75408706
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011539228.XA Pending CN112667791A (en) | 2020-12-23 | 2020-12-23 | Latent event prediction method, device, equipment and storage medium |
Country Status (2)
| Country | Link |
|---|---|
| CN (1) | CN112667791A (en) |
| WO (1) | WO2022134834A1 (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2022134834A1 (en) * | 2020-12-23 | 2022-06-30 | 深圳壹账通智能科技有限公司 | Potential event predicting method, apparatus and device, and storage medium |
| CN114780810A (en) * | 2022-04-22 | 2022-07-22 | 中国电信股份有限公司 | Data processing method, data processing device, storage medium and electronic equipment |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108733825A (en) * | 2018-05-23 | 2018-11-02 | 阿里巴巴集团控股有限公司 | A kind of objects trigger event prediction method and device |
| CN109918752A (en) * | 2019-02-26 | 2019-06-21 | 华南理工大学 | Mechanical failure diagnostic method, equipment and medium based on migration convolutional neural networks |
| CN109948851A (en) * | 2019-03-20 | 2019-06-28 | 北京百度网讯科技有限公司 | Method and apparatus for predicted events probability of happening |
| WO2019184217A1 (en) * | 2018-03-26 | 2019-10-03 | 平安科技（深圳）有限公司 | Hotspot event classification method and apparatus, and storage medium |
Family Cites Families (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110737758B (en) * | 2018-07-03 | 2022-07-05 | 百度在线网络技术（北京）有限公司 | Method and apparatus for generating a model |
| CN109508650A (en) * | 2018-10-23 | 2019-03-22 | 浙江农林大学 | A kind of wood recognition method based on transfer learning |
| CN112667791A (en) * | 2020-12-23 | 2021-04-16 | 深圳壹账通智能科技有限公司 | Latent event prediction method, device, equipment and storage medium |
-
2020
- 2020-12-23 CN CN202011539228.XA patent/CN112667791A/en active Pending
-
2021
- 2021-10-25 WO PCT/CN2021/126112 patent/WO2022134834A1/en active Application Filing
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2019184217A1 (en) * | 2018-03-26 | 2019-10-03 | 平安科技（深圳）有限公司 | Hotspot event classification method and apparatus, and storage medium |
| CN108733825A (en) * | 2018-05-23 | 2018-11-02 | 阿里巴巴集团控股有限公司 | A kind of objects trigger event prediction method and device |
| CN109918752A (en) * | 2019-02-26 | 2019-06-21 | 华南理工大学 | Mechanical failure diagnostic method, equipment and medium based on migration convolutional neural networks |
| CN109948851A (en) * | 2019-03-20 | 2019-06-28 | 北京百度网讯科技有限公司 | Method and apparatus for predicted events probability of happening |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2022134834A1 (en) * | 2020-12-23 | 2022-06-30 | 深圳壹账通智能科技有限公司 | Potential event predicting method, apparatus and device, and storage medium |
| CN114780810A (en) * | 2022-04-22 | 2022-07-22 | 中国电信股份有限公司 | Data processing method, data processing device, storage medium and electronic equipment |
| CN114780810B (en) * | 2022-04-22 | 2024-02-27 | 中国电信股份有限公司 | Data processing method and device, storage medium and electronic equipment |
Also Published As
| Publication number | Publication date |
|---|---|
| WO2022134834A1 (en) | 2022-06-30 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| WO2021159632A1 (en) | Intelligent questioning and answering method and apparatus, computer device, and computer storage medium | |
| CN108416032B (en) | Text classification method, device and storage medium | |
| US11270082B2 (en) | Hybrid natural language understanding | |
| CN115587175A (en) | Man-machine conversation and pre-training language model training method and system and electronic equipment | |
| CN113326374B (en) | Short text emotion classification method and system based on feature enhancement | |
| CN111930914A (en) | Question generation method and device, electronic equipment and computer-readable storage medium | |
| CN112185361B (en) | Voice recognition model training method and device, electronic equipment and storage medium | |
| CN117520523B (en) | Data processing method, device, equipment and storage medium | |
| WO2022134834A1 (en) | Potential event predicting method, apparatus and device, and storage medium | |
| CN111832305A (en) | User intention identification method, device, server and medium | |
| CN112528654A (en) | Natural language processing method and device and electronic equipment | |
| CN112101042A (en) | Text emotion recognition method and device, terminal device and storage medium | |
| CN115730597A (en) | Multi-level semantic intention recognition method and related equipment thereof | |
| CN114722176A (en) | Intelligent question answering method, device, medium and electronic equipment | |
| CN113987151A (en) | Intelligent customer service reply method and device | |
| CN110377706B (en) | Search sentence mining method and device based on deep learning | |
| US7783588B2 (en) | Context modeling architecture and framework | |
| CN115809658A (en) | Parallel corpus generation method and device and unsupervised synonymy transcription method and device | |
| US11636269B2 (en) | Content creation and prioritization | |
| CN114969347A (en) | Defect duplication checking implementation method and device, terminal equipment and storage medium | |
| CN113590768B (en) | Training method and device for text relevance model, question answering method and device | |
| CN110895924B (en) | Method and device for reading document content aloud, electronic equipment and readable storage medium | |
| Liu | Applying Natural Language Processing to Assessment | |
| CN117786416B (en) | Model training method, device, equipment, storage medium and product | |
| CN117851556A (en) | Question answering method, electronic equipment and computer storage medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| REG | Reference to a national code |
Ref country code: HK Ref legal event code: DE Ref document number: 40050554 Country of ref document: HK |
|
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20210416 |
|
| WD01 | Invention patent application deemed withdrawn after publication |