CN111782676A

CN111782676A - Probability prediction model training method, probability prediction method and probability prediction device

Info

Publication number: CN111782676A
Application number: CN202010399572.7A
Authority: CN
Inventors: 步佳昊; 杨扬; 李勇; 王金刚; 周翔; 张富峥; 陈�胜; 仙云森; 王仲远
Original assignee: Beijing Sankuai Online Technology Co Ltd
Current assignee: Beijing Sankuai Online Technology Co Ltd
Priority date: 2020-05-12
Filing date: 2020-05-12
Publication date: 2020-10-16

Abstract

The embodiment of the disclosure provides a probability prediction model training method, a probability prediction method and a probability prediction device. The probability prediction model training method comprises the following steps: acquiring a training sample associated with a business party; the training sample comprises an inquiry statement associated with the business party, a business party name corresponding to the business party, business party description information corresponding to the business party and an initial order associated with the inquiry statement and the business party; inputting query statements, the name of the service party and the description information of the service party into an initial probability prediction model to obtain the predicted lower single probability corresponding to the service party; calculating to obtain a loss value of the initial sequencing model according to the initial order and the predicted order placing probability; and under the condition that the loss value is within a preset range, taking the initial probability prediction model as a target probability prediction model. The method and the device for searching the business parties can improve the sequencing effect of the searched business parties, so that the searching result can better meet the requirements of users.

Description

Probability prediction model training method, probability prediction method and probability prediction device

Technical Field

Embodiments of the present disclosure relate to the field of model training technologies, and in particular, to a probabilistic predictive model training method, a probabilistic predictive device, an electronic device, and a computer-readable storage medium.

Background

The consortium search is a core entry of the consortium App (application) for bearing a user search request. Semantic relevance is one of the core factors in the cosmos core ranking that affect the user search experience. The American group search historically uses Query-doc (POI) text correlation features such as BM25, DSSM and the like, but the correlation of partial requests (such as merchant search, commodity search and the like) is not well solved. In the mei-qu search service, the merchant search accounts for about 30% of the total search traffic.

The text relevance feature is used for measuring the relevance between a user search keyword and a business (POI), and then is used as a feature to be fed into a fine-ranking model of an L2 layer to serve as the ranking of the search business. In a Mei-Tuan search scene, text relevance is calculated based on search keywords and merchant names to serve as one of important features of a core ranking model, a candidate merchant list is ranked, and the ranked candidate merchant list is fed back to a user.

The currently common text relevance feature calculation methods mainly include the following two methods:

a statistical-based method: including TF-IDF, BM25, etc. The method utilizes statistics to model the relevance probability of the words and the texts, and further calculates the relevance score, which is also two important text characteristics in the current American group search.

Secondly, a deep learning-based method: including DSSM, LSTM, BERT, etc. The method utilizes a deep neural network to learn the vector representation of the text, and obtains the relevance score by training a relevance classification model.

A statistical-based approach: the TF-IDF, BM25 and other methods are essentially bag-of-word models, which model single words/words, neglect the sequence relation between the words/words, and cannot obtain deeper semantic representation between query and poinamee. And the deep learning based method: BERT, et al, can learn the semantic representation between query and poinamee, but in the actual O2O scenario, it is not enough to judge the textual relevance between the user search term and the merchant just by query and poinamee. One practical example in a search is: the user searches for the 'Chinese water source takeaway', and recalled merchants comprise spa stores such as 'Chinese water margin'. The query that the user searches is very similar to the poinamee of the recalled merchant, but the merchant is not a matching category that the user wants to search. Another example is: the user searches for "2 to 3 party quiet" and the recalled merchants include the merchant named "23" which is again not the result intended by the user.

Disclosure of Invention

Embodiments of the present disclosure provide a probability prediction model training method, a probability prediction method, an electronic device, and a computer-readable storage medium, so as to improve a ranking effect of searched merchants, so that a search result better meets requirements of a user.

According to a first aspect of embodiments of the present disclosure, there is provided a probabilistic predictive model training method, including:

acquiring a training sample associated with a business party; the training sample comprises an inquiry statement associated with the business party, a business party name corresponding to the business party, business party description information corresponding to the business party and an initial order associated with the inquiry statement and the business party;

inputting query statements, the name of the service party and the description information of the service party into an initial probability prediction model to obtain the predicted lower single probability corresponding to the service party;

calculating to obtain a loss value of the initial sequencing model according to the initial order and the predicted order placing probability;

and under the condition that the loss value is within a preset range, taking the initial probability prediction model as a target probability prediction model.

Optionally, the initial probability prediction model includes an encoding layer and a probability prediction layer, and the inputting the query statement, the business party name, and the business party description information into the initial ranking model to obtain the predicted lower single probability corresponding to the business party includes:

splicing the query statement, the name of the service party and the description information of the service party to obtain a spliced text corresponding to the service party;

inputting the spliced text into the initial probability prediction model;

calling the coding layer to code the spliced text to generate a coding vector corresponding to the spliced text;

and calling the probability prediction layer to perform probability prediction on the training sample according to the coding vector, and determining the single probability under prediction.

Optionally, the splicing the query statement, the name of the service party and the description information of the service party to obtain a spliced text corresponding to the service party includes:

and connecting the query statement with the name of the service party, the name of the service party and the description information of the service party by adopting preset separators respectively to obtain the spliced text.

Optionally, the invoking the coding layer to code the spliced text to generate a coding vector corresponding to the spliced text includes:

and calling the coding layer to code the spliced text to obtain a word coding vector, a segment coding vector and a position coding vector corresponding to the spliced text.

Optionally, the invoking the probability prediction layer to perform probability prediction on the training sample according to the coding vector, and determining the single probability under prediction includes:

and calling the probability prediction layer to process the word coding vector, the segment coding vector and the position coding vector to generate the single probability under prediction.

Optionally, the service party description information includes: at least one of service party category information, service party tag information, and service party location information.

According to a second aspect of embodiments of the present disclosure, there is provided a probability prediction method including:

acquiring a query statement;

inquiring according to the inquiry statement to obtain at least one business party matched with the inquiry statement;

acquiring a service party name and service party description information corresponding to the at least one service party;

inputting the query statement, the name of the business party and the description information of the business party into a target probability prediction model;

processing the query statement, the name of the business party and the description information of the business party through the target probability prediction model, and determining the ordering probability corresponding to the at least one business party;

and sequencing the at least one service party according to the ordering probability, and sending the sequenced service parties to a terminal corresponding to the query statement.

Optionally, the inputting the query statement, the business party name, and the business party description information into a target probability prediction model includes:

and inputting the spliced text into the target probability prediction model.

Optionally, the target probability prediction model includes an encoding layer and a probability prediction layer, and the determining, by processing the query statement, the business party name, and the business party description information through the target probability prediction model, a lower order probability corresponding to the at least one business party includes:

calling the coding layer to code the query statement, the name of the service party and the description information of the service party to generate a corresponding coding vector;

and calling the probability prediction layer to process the coding vector to generate a ordering probability corresponding to the at least one service party.

Optionally, the sorting the at least one service party according to the ordering probability and sending the sorted service party to the terminal corresponding to the query statement includes:

sequencing the at least one service party according to the sequence of the ordering probability from big to small to obtain sequenced service parties;

and sending the sequenced service parties to the terminal.

According to a third aspect of embodiments of the present disclosure, there is provided a probabilistic predictive model training device including:

the training sample acquisition module is used for acquiring a training sample associated with a business party; the training sample comprises an inquiry statement associated with the business party, a business party name corresponding to the business party, business party description information corresponding to the business party and an initial order associated with the inquiry statement and the business party;

the prediction probability obtaining module is used for inputting the query statement, the name of the service party and the description information of the service party into an initial probability prediction model and obtaining the prediction lower single probability corresponding to the service party;

the loss value calculation module is used for calculating the loss value of the initial sequencing model according to the initial order and the predicted order placing probability;

and the target probability model acquisition module is used for taking the initial probability prediction model as a target probability prediction model under the condition that the loss value is within a preset range.

Optionally, the initial probabilistic prediction model includes an encoding layer and a probability prediction layer, and the prediction probability obtaining module includes:

the first splicing text acquisition unit is used for splicing the query statement, the name of the service party and the description information of the service party to obtain a splicing text corresponding to the service party;

a first spliced text input unit, configured to input the spliced text to the initial probability prediction model;

the first coding vector generating unit is used for calling the coding layer to code the spliced text and generating a coding vector corresponding to the spliced text;

and the first lower single probability determining unit is used for calling the probability prediction layer to carry out probability prediction on the training sample according to the coding vector and determining the prediction lower single probability.

Optionally, the first stitched text acquiring unit includes:

and the splicing text obtaining subunit is used for connecting the query statement with the name of the service party, the name of the service party and the description information of the service party by adopting preset separators respectively to obtain the splicing text.

Optionally, the first encoding vector generating unit includes:

and the coding vector acquisition subunit is used for calling the coding layer to code the spliced text to obtain a word coding vector, a segment coding vector and a position coding vector corresponding to the spliced text.

Optionally, the first lower order probability determination unit includes:

and the lower single probability prediction generation subunit is used for calling the probability prediction layer to process the word coding vector, the segment coding vector and the position coding vector to generate the lower single probability prediction.

According to a fourth aspect of embodiments of the present disclosure, there is provided a probability prediction apparatus including:

a query statement acquisition module for acquiring a query statement;

the business party acquisition module is used for inquiring according to the inquiry statement to obtain at least one business party matched with the inquiry statement;

the service party information acquisition module is used for acquiring the name of a service party and the description information of the service party corresponding to the at least one service party;

the business party information input module is used for inputting the query statement, the business party name and the business party description information into a target probability prediction model;

the ordering probability determining module is used for processing the query statement, the name of the business party and the description information of the business party through the target probability prediction model and determining the ordering probability corresponding to the at least one business party;

and the service party sending module is used for sequencing the at least one service party according to the ordering probability and sending the sequenced service party to the terminal corresponding to the query statement.

Optionally, the service party information input module includes:

a second splicing text obtaining unit, configured to splice the query statement, the name of the service party, and the description information of the service party to obtain a splicing text corresponding to the service party;

and the second spliced text input unit is used for inputting the spliced text into the target probability prediction model.

Optionally, the target probability prediction model includes an encoding layer and a probability prediction layer, and the lower single probability determination module includes:

the second coding vector generating unit is used for calling the coding layer to code the query statement, the business party name and the business party description information to generate a corresponding coding vector;

and the ordering probability generating unit is used for calling the probability prediction layer to process the coding vector and generating the ordering probability corresponding to the at least one service party.

Optionally, the service sending module includes:

the service party sequencing unit is used for sequencing the at least one service party according to the sequence of the ordering probability from large to small to obtain the sequenced service parties;

and the service party sending unit is used for sending the sequenced service parties to the terminal.

According to a fifth aspect of embodiments of the present disclosure, there is provided an electronic apparatus including:

a processor, a memory, and a computer program stored on the memory and executable on the processor, the processor implementing any of the above probabilistic predictive model training methods, or any of the above probabilistic predictive methods, when executing the program.

According to a sixth aspect of embodiments of the present disclosure, there is provided a readable storage medium, wherein instructions, when executed by a processor of an electronic device, enable the electronic device to perform the probabilistic predictive model training method according to any of the above, or the probabilistic predictive method according to any of the above.

According to the scheme provided by the embodiment of the disclosure, a training sample associated with a business party is obtained, the training sample comprises an inquiry statement associated with the business party, a business party name corresponding to the business party, business party description information corresponding to the business party and an initial order associated with the inquiry statement and the business party, the inquiry statement, the business party name and the business party description information are input into an initial probability prediction model, a predicted ordering probability corresponding to the business party is obtained, a loss value of the initial sequencing model is obtained through calculation according to the initial order and the predicted ordering probability, and the initial probability prediction model is used as a target probability prediction model under the condition that the loss value is within a preset range. According to the method and the device, the probability prediction model is improved, the description information of the business party is introduced, the sequencing effect of the searched business party is improved to a certain extent, and the searching result is more in line with the requirements of the user.

Drawings

In order to more clearly illustrate the technical solutions of the embodiments of the present disclosure, the drawings needed to be used in the description of the embodiments of the present disclosure will be briefly introduced below, and it is obvious that the drawings in the following description are only some embodiments of the present disclosure, and it is obvious for those skilled in the art that other drawings can be obtained according to the drawings without inventive exercise.

FIG. 1 is a flowchart illustrating steps of a probabilistic predictive model training method provided by an embodiment of the present disclosure;

FIG. 1a shows a schematic diagram of a bert model in the prior art;

FIG. 1b is a schematic diagram illustrating a text encoding provided by an embodiment of the present disclosure;

FIG. 2 is a flow chart illustrating the steps of a probabilistic prediction method provided by an embodiment of the present disclosure;

FIG. 3 is a schematic structural diagram of a probabilistic predictive model training device provided by an embodiment of the present disclosure;

fig. 4 shows a schematic structural diagram of a probability prediction apparatus provided by an embodiment of the present disclosure.

Detailed Description

Technical solutions in the embodiments of the present disclosure will be clearly and completely described below with reference to the drawings in the embodiments of the present disclosure, and it is apparent that the described embodiments are some, but not all, of the embodiments of the present disclosure. All other embodiments, which can be obtained by a person skilled in the art without making creative efforts based on the embodiments of the present disclosure, belong to the protection scope of the embodiments of the present disclosure.

Example one

Referring to fig. 1, a flowchart illustrating steps of a probabilistic predictive model training method provided in an embodiment of the present disclosure is shown, where the probabilistic predictive model training method may specifically include the following steps:

step 101: acquiring a training sample associated with a business party; the training sample comprises an inquiry statement associated with the business party, a business party name corresponding to the business party, business party description information corresponding to the business party, and an initial order associated with the inquiry statement and the business party.

Embodiments of the present disclosure may be applied in a scenario where a probabilistic predictive model is trained.

The probability prediction model is used for performing single probability prediction on a business party queried by a query statement.

The business party refers to a party capable of generating an order, such as a hotel, a restaurant and the like, and specifically, the business party may be determined according to actual situations.

The training sample refers to a sample for training the probability prediction model, and the training sample may include a query statement associated with a business party, a business party name of the business party, business party description information of the business party, and the like.

In this embodiment, the service party description information may include at least one of service party category information, service party tag information, and service party location information.

The business party category information refers to a category corresponding to an order provided by a business party, for example, business parties take merchants as an example, and the American group side classifies all merchants in detail. The classification system comprises 3 levels of categories, wherein the category of level 1 comprises hotel, education and training and the like, the category of level 2 comprises four-star/high-grade, and teaching aid for ascending education and the like, and the category of level 3 comprises high-grade, examination and the like.

The service party label information refers to a label given to the service party according to information such as user comments, for example, the service party takes a merchant as an example, and the NLP (natural language processing) center digs a large number of text labels for the merchant from the user comments based on methods such as LDA (document theme generation model), rules, relationship extraction and the like, such as "suitable for a baby", "elegant in classical", "open air position" and the like. For broad search terms (e.g., "eat in the open air", etc.).

The service location information refers to a geographic location where the service party is located, for example, the service party takes a business as an example, and the business is located in a city, a street, and the like.

In the training sample, each query statement and the business side correspond to an initial order, namely whether to place an order when the business side queries by using the query statement includes two cases: 1. and (3) ordering, 2 not ordering, in the embodiment, the query sentences which are ordered all can be selected to train the model.

After the training samples associated with the business party are obtained, step 102 is performed.

Step 102: and inputting the query statement, the name of the service party and the description information of the service party into an initial probability prediction model to obtain the predicted lower single probability corresponding to the service party.

The initial probabilistic prediction model is a probabilistic prediction model that has not been trained, and the training samples obtained in the above steps are used to train the initial probabilistic prediction model.

The predicted ordering probability refers to the ordering probability corresponding to the business party predicted by the initial probability prediction model after the query statement, the business party name and the business party description information are input to the initial probability model through the initial probability prediction model, namely the probability that an order can be generated when the query statement is input.

After the training sample associated with the business party is obtained, the query statement, the name of the business party, and the description information of the business party included in the training sample may be input to the initial probability prediction model together, and the query statement, the name of the business party, and the description information of the business party are processed through the initial probability prediction model to obtain the predicted lower single probability corresponding to the business party, which may be described in detail in combination with the following specific implementation manner.

In a specific implementation manner of the present disclosure, the step 102 may include:

substep A1: and splicing the query statement, the name of the service party and the description information of the service party to obtain a spliced text corresponding to the service party.

In this embodiment, the splicing text refers to a text obtained by splicing the query statement, the name of the business party, and the description information of the business party, and specifically, corresponding separators may be added between the query statement and the name of the business party, and between the name of the business party and the description information of the business party, so as to splice these three kinds of information, for example, as shown in fig. 1b, the input query statement, the name of the business party, and the description information of the business party (the type of the business party, the label of the business party, etc.) may be spliced by using a separator [ SEP ] to obtain the splicing text. Of course, the two delimiters may be the same or different, and may be determined according to the service requirement.

After the spliced text is obtained, sub-step A2 is performed.

Substep A2: inputting the stitched text to the initial probabilistic predictive model.

After the stitched text is obtained, the stitched text may then be input to the initial probabilistic predictive model and, in turn, sub-step a3 is performed.

Substep A3: and calling the coding layer to code the spliced text to generate a coding vector corresponding to the spliced text.

In this embodiment, the initial probability prediction model may include an encoding layer and a probability prediction layer, where the encoding layer may be configured to encode the spliced text to generate an encoding vector corresponding to the spliced text, and the probability prediction layer may be configured to perform prediction according to the encoding vector to obtain a ordering probability corresponding to the service party by prediction.

After the spliced text is input into the initial probability model, the coding layer may be invoked to process the spliced text to generate a coding vector corresponding to the spliced text. In particular, the encoding vectors may include word encoding vectors, segment encoding vectors, and position encoding vectors.

The word encoding vector refers to an encoding vector corresponding to each word in the input spliced text, such as the word encoding shown in fig. 1 b.

The segment coding vector is a vector corresponding to a segment to which each word belongs in the spliced text, and in the original bert model, only two segments (as shown in fig. 1 a) are provided, namely, a query statement query and a frequency band corresponding to a service party name.

The position coding vector refers to a vector corresponding to the position of each word in the spliced text. For a position-encoded vector: the query sentence, the name of the service party and the description information of the service party are all independent sentences with realistic meanings, so the position code is unchanged, and when the description information of the service party contains category information (namely the information of the class of the service party), the category information is spliced by 3-level category labels and does not form an independent sentence, so the position code is directly removed, and the language sequence concept without sentence level is represented.

After the code vector corresponding to the spliced text is obtained, sub-step a4 is performed.

Substep A4: and calling the probability prediction layer to perform probability prediction on the training sample according to the coding vector, and determining the single probability under prediction.

After the coding vector corresponding to the spliced text is obtained, a probability prediction layer can be called to predict the training sample according to the coding vector so as to obtain the single probability under prediction.

After the predicted next single probability is obtained, step 103 is executed.

Step 103: and calculating to obtain a loss value of the initial sequencing model according to the initial order and the predicted order placing probability.

The loss value may represent a deviation degree between the initial order and the predicted order placing probability, and specifically, a difference between the initial order and the predicted order placing probability may be calculated, and then, the deviation degree between the initial order and the predicted order placing probability may be calculated according to the difference between the initial order and the predicted order placing probability.

After the loss value is obtained, step 104 is performed.

Step 104: and under the condition that the loss value is within a preset range, taking the initial probability prediction model as a target probability prediction model.

In this embodiment, the preset range may be preset by a research and development staff according to an actual application scenario and an actual requirement, and the specific numerical value of the preset range is not limited in the embodiment of the present invention.

If the loss value is within the preset range, the deviation between the initial order associated with each query statement and the business party and the corresponding predicted order-placing probability is very small, at this time, the probability prediction model can be considered to be capable of accurately predicting the order-placing probability of the query statement corresponding to the queried business party, correspondingly, the initial probability prediction model can be used as a target probability prediction model, and the target probability prediction model can perform subsequent probability prediction.

If the loss value is out of the preset range, the deviation between the initial order associated with each query statement and the business party and the corresponding predicted order taking probability is large, at the moment, the number of training samples can be increased, and the initial probability prediction model is continuously trained, so that the finally obtained loss value is in the preset range.

The probability prediction model training method provided by the embodiment of the disclosure includes the steps of obtaining a training sample associated with a business party, wherein the training sample comprises an inquiry statement associated with the business party, a business party name corresponding to the business party, business party description information corresponding to the business party, and an initial order associated with the inquiry statement and the business party, inputting the inquiry statement, the business party name and the business party description information into an initial probability prediction model, obtaining a predicted ordering probability corresponding to the business party, calculating a loss value of the initial ranking model according to the initial order and the predicted ordering probability, and taking the initial probability prediction model as a target probability prediction model under the condition that the loss value is within a preset range. According to the method and the device, the probability prediction model is improved, the description information of the business party is introduced, the sequencing effect of the searched business party is improved to a certain extent, and the searching result is more in line with the requirements of the user.

Example two

Referring to fig. 2, a flowchart illustrating steps of a probability prediction method provided by an embodiment of the present disclosure is shown, where the probability prediction method specifically includes the following steps:

step 201: a query statement is obtained.

In this embodiment, the query statement refers to a statement for querying a relevant business party.

In some examples, the query statement may be a statement input by a user in a specific APP, for example, when the user wants to search a certain business, the user may input query information in a search box corresponding to the APP, and then the query information input by the user may be used as the query statement.

In some examples, the query statement may also be a statement obtained in the internet for querying a business party.

It is to be understood that the above-mentioned examples are only examples set forth for better understanding of the technical solutions of the embodiments of the present disclosure, and are not to be taken as the only limitation of the present disclosure.

After the query statement is obtained, step 202 is performed.

Step 202: and querying according to the query statement to obtain at least one business party matched with the query statement.

After the query statement is obtained, information query can be performed in the database according to the query statement to obtain at least one service party matched with the query statement.

After at least one business party matching the query statement is obtained, step 203 is performed.

Step 203: and acquiring the name and the description information of the service party corresponding to the at least one service party.

The name of the service party refers to the name of the service party, for example, the service party takes a merchant as an example, and the name of the service party is the name of the merchant.

The service party description information refers to information capable of describing a service party, and in this embodiment, the service party description information may include: at least one of service party item information, service party label information, service party position information and the like.

The service party label information refers to a label given to the service party according to information such as user comments, for example, the service party takes a merchant as an example, and the NLP (natural Language Processing) center digs a large number of text labels for the merchant from the user comments based on methods such as LDA (document theme generation model), rules, relationship extraction and the like, such as "suitable for a baby", "classic elegance", "open air position" and the like. For broad search terms (e.g., "eat in the open air", etc.).

After the at least one service party matched with the query statement is obtained, a service party name and service party description information corresponding to the at least one service party may be obtained, and then step 204 is performed.

Step 204: and inputting the query statement, the name of the service party and the description information of the service party into a target probability prediction model.

After the business party name and the business party description information of at least one business party are obtained, the query statement, the business party name and the business party description information can be input into the target probability prediction model, and then the ordering probability of the query statement relative to at least one business party can be predicted through the target probability prediction model.

In this embodiment, the query statement, the name of the service party, and the description information of the service party may be first spliced, so that the spliced text is input to the target probability prediction model, and specifically, the following specific implementation manner may be combined for detailed description.

In a specific implementation manner of the present disclosure, the step 204 may include:

substep B1: and splicing the query statement, the name of the service party and the description information of the service party to obtain a spliced text corresponding to the service party.

Substep B2: and inputting the spliced text into the target probability prediction model.

After the spliced text is obtained, the spliced text can be input into the target probability model, and then the target probability model predicts the ordering probability corresponding to at least one service party according to the spliced text.

After the query statement, business party name and business party description information are input to the target probability prediction model, step 205 is performed.

Step 205: and processing the query statement, the name of the service party and the description information of the service party through the target probability prediction model, and determining the ordering probability corresponding to the at least one service party.

The ordering probability refers to the probability of ordering at least one service party when at least one service party is obtained by query according to the query statement.

After the query statement, the business party name and the business party description information are input into the target probability prediction model, the query statement, the business party name and the business party description information can be processed through the target probability prediction model to determine the ordering probability corresponding to at least one business party. In particular, the detailed description may be combined with the following specific implementations.

In a specific implementation manner of the present disclosure, the step 205 may include:

substep C1: and calling the coding layer to code the query statement, the name of the service party and the description information of the service party to generate a corresponding coding vector.

In this embodiment, the target probability prediction model may include an encoding layer and a probability prediction layer, where the encoding layer may be configured to encode the spliced text to generate an encoding vector corresponding to the spliced text, and the probability prediction layer may be configured to perform prediction according to the encoding vector to obtain a ordering probability corresponding to the service party by prediction.

After the coding layer is called to perform coding processing on the query statement, the business party name and the business party description information and generate corresponding coding vectors, sub-step C2 is performed.

Substep C2: and calling the probability prediction layer to process the coding vector to generate a ordering probability corresponding to the at least one service party.

After the code vector corresponding to the spliced text is obtained, the probability prediction layer can be called to process the code vector according to the code vector so as to obtain the ordering probability of at least one service party.

After generating the ordering probability for at least one of the traffic parties, step 206 is performed.

Step 206: and sequencing the at least one service party according to the ordering probability, and sending the sequenced service parties to a terminal corresponding to the query statement.

After the ordering probability of at least one service party is obtained, the at least one service party can be ranked according to the ordering probability and the ranked service party is sent to the terminal corresponding to the query statement, specifically, the at least one service party can be ranked according to the descending probability from large to small, and the ranked service party is sent to the terminal corresponding to the query statement, namely, the closer to the service party, the higher the ordering probability of the user is, so that the ordering probability of the user can be improved, and the search result can better meet the user requirement.

The probability prediction method provided by the embodiment of the disclosure obtains at least one business party matched with an inquiry statement by obtaining the inquiry statement and inquiring according to the inquiry statement, obtains a business party name and business party description information corresponding to the at least one business party, inputs the inquiry statement, the business party name and the business party description information into a target probability prediction model, processes the inquiry statement, the business party name and the business party description information through the target probability prediction model, determines an ordering probability corresponding to the at least one business party, orders the at least one business party according to the ordering probability, and sends the ordered business party to a terminal corresponding to the inquiry statement. According to the method and the device, the searched business parties are ranked through the pre-trained probability prediction model, the names of the business parties are combined, and the description information of the business parties is also combined, so that the ranking effect of the searched business parties can be improved to a certain extent, and the search result can better meet the requirements of users.

EXAMPLE III

Referring to fig. 3, a schematic structural diagram of a probabilistic predictive model training device provided in an embodiment of the present disclosure is shown, where the probabilistic predictive model training device may specifically include the following modules:

a training sample obtaining module 310, configured to obtain a training sample associated with a service party; the training sample comprises an inquiry statement associated with the business party, a business party name corresponding to the business party, business party description information corresponding to the business party and an initial order associated with the inquiry statement and the business party;

a prediction probability obtaining module 320, configured to input the query statement, the name of the service party, and the description information of the service party to an initial probability prediction model, and obtain a predicted lower single probability corresponding to the service party;

a loss value calculation module 330, configured to calculate a loss value of the initial ordering model according to the initial order and the predicted order placing probability;

and a target probability model obtaining module 340, configured to take the initial probability prediction model as a target probability prediction model when the loss value is within a preset range.

Optionally, the initial probability prediction model includes an encoding layer and a probability prediction layer, and the prediction probability obtaining module 320 includes:

Optionally, the first stitched text acquiring unit includes:

Optionally, the first encoding vector generating unit includes:

Optionally, the first lower order probability determination unit includes:

The probability prediction model training device provided by the embodiment of the disclosure inputs query statement, business side name and business side description information into an initial probability prediction model by acquiring a training sample associated with a business side, wherein the training sample comprises a query statement associated with the business side, a business side name corresponding to the business side, business side description information corresponding to the business side and an initial order associated with the query statement and the business side, acquires a predicted ordering probability corresponding to the business side, calculates a loss value of the initial ranking model according to the initial order and the predicted ordering probability, and takes the initial probability prediction model as a target probability prediction model under the condition that the loss value is within a preset range. According to the method and the device, the probability prediction model is improved, the description information of the business party is introduced, the sequencing effect of the searched business party is improved to a certain extent, and the searching result is more in line with the requirements of the user.

Example four

Referring to fig. 4, a schematic structural diagram of a probability prediction apparatus provided in an embodiment of the present disclosure is shown, where the probability prediction apparatus may specifically include the following modules:

a query statement obtaining module 410, configured to obtain a query statement;

a service party obtaining module 420, configured to perform query according to the query statement, so as to obtain at least one service party matched with the query statement;

a service party information obtaining module 430, configured to obtain a service party name and service party description information corresponding to the at least one service party;

a service party information input module 440, configured to input the query statement, the name of the service party, and the description information of the service party to a target probability prediction model;

the ordering probability determining module 450 is configured to process the query statement, the name of the business party and the description information of the business party through the target probability prediction model, and determine an ordering probability corresponding to the at least one business party;

and the service party sending module 460 is configured to rank the at least one service party according to the ordering probability, and send the ranked service party to the terminal corresponding to the query statement.

Optionally, the service party information input module 430 includes:

Optionally, the target probability prediction model includes an encoding layer and a probability prediction layer, and the lower single probability determining module 450 includes:

Optionally, the service side sending module 460 includes:

The probability prediction device provided by the embodiment of the disclosure obtains at least one service party matched with an inquiry statement by obtaining the inquiry statement and inquiring according to the inquiry statement, obtains a service party name and service party description information corresponding to the at least one service party, inputs the inquiry statement, the service party name and the service party description information into a target probability prediction model, processes the inquiry statement, the service party name and the service party description information through the target probability prediction model, determines an ordering probability corresponding to the at least one service party, orders the at least one service party according to the ordering probability, and sends the ordered service party to a terminal corresponding to the inquiry statement. According to the method and the device, the searched business parties are ranked through the pre-trained probability prediction model, the names of the business parties are combined, and the description information of the business parties is also combined, so that the ranking effect of the searched business parties can be improved to a certain extent, and the search result can better meet the requirements of users.

An embodiment of the present disclosure also provides an electronic device, including: a processor, a memory, and a computer program stored on the memory and executable on the processor, the processor implementing the probabilistic predictive model training method of the foregoing embodiment, or the probabilistic predictive method of the foregoing embodiment, when executing the program.

Embodiments of the present disclosure also provide a computer-readable storage medium, in which instructions, when executed by a processor of an electronic device, enable the electronic device to perform the probabilistic predictive model training method of the foregoing embodiments, or the probabilistic predictive method of the foregoing embodiments.

For the device embodiment, since it is basically similar to the method embodiment, the description is simple, and for the relevant points, refer to the partial description of the method embodiment.

The algorithms and displays presented herein are not inherently related to any particular computer, virtual machine, or other apparatus. Various general purpose systems may also be used with the teachings herein. The required structure for constructing such a system will be apparent from the description above. In addition, embodiments of the present disclosure are not directed to any particular programming language. It is appreciated that a variety of programming languages may be used to implement the teachings of the embodiments of the present disclosure as described herein, and any descriptions of specific languages are provided above to disclose the best modes of the embodiments of the present disclosure.

In the description provided herein, numerous specific details are set forth. However, it is understood that embodiments of the disclosure may be practiced without these specific details. In some instances, well-known methods, structures and techniques have not been shown in detail in order not to obscure an understanding of this description.

Similarly, it should be appreciated that in the foregoing description of exemplary embodiments of the disclosure, various features of the embodiments of the disclosure are sometimes grouped together in a single embodiment, figure, or description thereof for the purpose of streamlining the disclosure and aiding in the understanding of one or more of the various inventive aspects. However, the disclosed method should not be interpreted as reflecting an intention that: that is, claimed embodiments of the disclosure require more features than are expressly recited in each claim. Rather, as the following claims reflect, inventive aspects lie in less than all features of a single foregoing disclosed embodiment. Thus, the claims following the detailed description are hereby expressly incorporated into this detailed description, with each claim standing on its own as a separate embodiment of an embodiment of this disclosure.

Those skilled in the art will appreciate that the modules in the device in an embodiment may be adaptively changed and disposed in one or more devices different from the embodiment. The modules or units or components of the embodiments may be combined into one module or unit or component, and furthermore they may be divided into a plurality of sub-modules or sub-units or sub-components. All of the features disclosed in this specification (including any accompanying claims, abstract and drawings), and all of the processes or elements of any method or apparatus so disclosed, may be combined in any combination, except combinations where at least some of such features and/or processes or elements are mutually exclusive. Each feature disclosed in this specification (including any accompanying claims, abstract and drawings) may be replaced by alternative features serving the same, equivalent or similar purpose, unless expressly stated otherwise.

The various component embodiments of the disclosure may be implemented in hardware, or in software modules running on one or more processors, or in a combination thereof. It will be understood by those skilled in the art that a microprocessor or Digital Signal Processor (DSP) may be used in practice to implement some or all of the functions of some or all of the components in a motion picture generating device according to an embodiment of the present disclosure. Embodiments of the present disclosure may also be implemented as an apparatus or device program for performing a portion or all of the methods described herein. Such programs implementing embodiments of the present disclosure may be stored on a computer readable medium or may be in the form of one or more signals. Such a signal may be downloaded from an internet website or provided on a carrier signal or in any other form.

It should be noted that the above-mentioned embodiments illustrate rather than limit embodiments of the disclosure, and that those skilled in the art will be able to design alternative embodiments without departing from the scope of the appended claims. In the claims, any reference signs placed between parentheses shall not be construed as limiting the claim. The word "comprising" does not exclude the presence of elements or steps not listed in a claim. The word "a" or "an" preceding an element does not exclude the presence of a plurality of such elements. Embodiments of the disclosure may be implemented by means of hardware comprising several distinct elements, and by means of a suitably programmed computer. In the unit claims enumerating several means, several of these means may be embodied by one and the same item of hardware. The usage of the words first, second and third, etcetera do not indicate any ordering. These words may be interpreted as names.

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 above description is only for the purpose of illustrating the preferred embodiments of the present disclosure and is not to be construed as limiting the embodiments of the present disclosure, and any modifications, equivalents, improvements and the like that are made within the spirit and principle of the embodiments of the present disclosure are intended to be included within the scope of the embodiments of the present disclosure.

The above description is only a specific implementation of the embodiments of the present disclosure, but the scope of the embodiments of the present disclosure is not limited thereto, and any person skilled in the art can easily conceive of changes or substitutions within the technical scope of the embodiments of the present disclosure, and all the changes or substitutions should be covered by the scope of the embodiments of the present disclosure. Therefore, the protection scope of the embodiments of the present disclosure shall be subject to the protection scope of the claims.

Claims

1. A probabilistic predictive model training method, comprising:

2. The method of claim 1, wherein the initial probabilistic prediction model includes an encoding layer and a probabilistic prediction layer, and the inputting the query statement, the business party name and the business party description information into the initial ranking model to obtain the predicted lower single probability corresponding to the business party comprises:

inputting the spliced text into the initial probability prediction model;

3. The method of claim 2, wherein the splicing the query statement, the name of the service party and the description information of the service party to obtain a spliced text corresponding to the service party comprises:

4. The method according to claim 2, wherein the invoking the coding layer to code the spliced text to generate a coding vector corresponding to the spliced text comprises:

5. The method of claim 4, wherein invoking the probabilistic prediction layer to probabilistically predict the training samples according to the coding vector, and wherein determining the prediction singleton probability comprises:

6. The method of claim 1, wherein the service party description information comprises: at least one of service party category information, service party tag information, and service party location information.

7. A method of probability prediction, comprising:

acquiring a query statement;

8. The method of claim 7, wherein inputting the query statement, the business party name, and the business party description information to a target probability prediction model comprises:

and inputting the spliced text into the target probability prediction model.

9. The method of claim 7, wherein the target probabilistic prediction model comprises an encoding layer and a probabilistic prediction layer, and wherein the determining the ordering probability corresponding to the at least one service party by processing the query statement, the service party name and the service party description information through the target probabilistic prediction model comprises:

10. The method according to claim 7, wherein the sorting the at least one service party according to the ordering probability and sending the sorted service party to the terminal corresponding to the query statement comprises:

and sending the sequenced service parties to the terminal.

11. A probabilistic predictive model training device, comprising:

12. The apparatus of claim 11, wherein the initial probabilistic predictive model comprises a coding layer and a probabilistic predictive layer, and wherein the prediction probability obtaining module comprises:

13. The apparatus of claim 12, wherein the first stitched text obtaining unit comprises:

14. The apparatus of claim 12, wherein the first code vector generation unit comprises:

15. The apparatus of claim 14, wherein the first lower order probability determination unit comprises:

16. The apparatus of claim 11, wherein the service party description information comprises: at least one of service party category information, service party tag information, and service party location information.

17. A probability prediction apparatus, comprising:

a query statement acquisition module for acquiring a query statement;

18. The apparatus of claim 17, wherein the service information input module comprises:

19. The apparatus of claim 17, wherein the target probabilistic prediction model comprises a coding layer and a probabilistic prediction layer, and wherein the lower single probability determination module comprises:

20. The apparatus of claim 17, wherein the server sending module comprises:

21. An electronic device, comprising:

a processor, a memory, and a computer program stored on the memory and executable on the processor, the processor implementing the probabilistic predictive model training method of any of claims 1 to 6, or the probabilistic predictive method of any of claims 7 to 10 when executing the program.

22. A readable storage medium, wherein instructions in the storage medium, when executed by a processor of an electronic device, enable the electronic device to perform the probabilistic predictive model training method of any of claims 1 to 6, or the probabilistic predictive method of any of claims 7 to 10.