CN113239175A

CN113239175A - Method for displaying candidate sentence list and terminal equipment

Info

Publication number: CN113239175A
Application number: CN202110651345.3A
Authority: CN
Inventors: 马建
Original assignee: Ping An Life Insurance Company of China Ltd
Current assignee: Ping An Life Insurance Company of China Ltd
Priority date: 2021-06-10
Filing date: 2021-06-10
Publication date: 2021-08-10

Abstract

The application relates to the technical field of big data, and provides a method for displaying a candidate sentence list, a device for displaying the candidate sentence list, a terminal device and a computer-readable storage medium. When a user triggers a preset recall operation through a session interface, calling a plurality of preset recall modules to respectively perform the candidate statement recall operation according to information input by the user on the session interface to obtain a plurality of candidate statements, encoding characteristic information of each candidate statement according to a preset encoding strategy to obtain statement characteristic vectors, inputting the statement characteristic vectors corresponding to each candidate statement into a pre-trained statement sequencing model to obtain a sequencing value of each candidate statement, and displaying the candidate statement list obtained by sequencing according to the sequencing value of each candidate statement in the session interface, so that the candidate statements are sequenced by using uniform sequencing logic, and the applicability of the candidate statement list is improved.

Description

Method for displaying candidate sentence list and terminal equipment

Technical Field

The invention belongs to the technical field of big data, and particularly relates to a method for displaying a candidate sentence list, a device for displaying the candidate sentence list, a terminal device and a computer-readable storage medium.

Background

At present, with the development of the field of artificial intelligence, an intelligent conversation system developed based on an artificial intelligence technology is widely applied to various scenes. Such as: intelligent customer service, intelligent chat robot, intelligent voice assistant, etc. Taking the intelligent customer service as an example, when a user makes a conversation with an intelligent client through a conversation interface on a terminal, and detects voice or characters input by the user through the conversation interface, the candidate sentence recall operation can be performed based on the voice content or the character content, and then a candidate sentence list obtained by the candidate sentence recall operation is displayed, so that the user can conveniently select a target candidate sentence from the candidate sentence list.

However, when the candidate sentence recall operation is performed, different recall modules are required to be used to recall the candidate question sentences according to different service logics or different recall channels respectively to obtain a plurality of candidate sentences, and then the ranking operation among the candidate sentences is realized based on the scores of the candidate sentences by the recall modules. However, because different recall modules give different scoring systems, and recalled candidate sentences have diversified contents, for example, the recalled candidate sentences may include texts, pictures, audio, video, various cards, and the like, the different recall modules score a plurality of candidate sentences, and the scores of the obtained candidate sentences are not comparable, so that the candidate sentence list obtained by ranking according to the scores of the candidate sentences has low applicability.

Disclosure of Invention

In view of this, embodiments of the present application provide a method for displaying a candidate sentence list, an apparatus for displaying a candidate sentence list, a terminal device, and a computer-readable storage medium, so as to solve the problem that the existing candidate sentence list is low in applicability.

A first aspect of an embodiment of the present application provides a method for displaying a candidate sentence list, including:

responding to a preset recall operation triggered by a user through a session interface, calling a plurality of preset recall modules to respectively perform candidate statement recall operation according to information input by the user on the session interface to obtain a plurality of candidate statements;

coding the feature information of each candidate statement according to a preset coding strategy to obtain statement feature vectors;

inputting the sentence characteristic vector into a pre-trained sentence sequencing model to obtain a sequencing value of each candidate sentence; wherein the ranking value is used to indicate a high or low quality of each of the candidate sentences;

and displaying a candidate sentence list obtained by sequencing the candidate sentences according to the sequencing value of each candidate sentence in the conversation interface.

A second aspect of an embodiment of the present application provides an apparatus for displaying a candidate sentence list, including:

the system comprises a recall unit, a recall unit and a recall unit, wherein the recall unit is used for responding to a preset recall operation triggered by a user through a session interface, calling a plurality of preset recall modules to respectively perform candidate statement recall operation according to information input by the user on the session interface, and obtaining a plurality of candidate statements;

the encoding unit is used for encoding the feature information of each candidate statement according to a preset encoding strategy to obtain statement feature vectors;

the sorting unit is used for inputting the statement feature vectors into a pre-trained statement sorting model to obtain a sorting value of each candidate statement; wherein the ranking value is used to indicate a high or low quality of each of the candidate sentences;

and the display unit is used for displaying a candidate sentence list obtained by sequencing the candidate sentences according to the sequencing value of each candidate sentence in the conversation interface.

A third aspect of embodiments of the present application provides a terminal device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, where the processor implements the steps of the first aspect when executing the computer program.

A fourth aspect of embodiments of the present application provides a computer-readable storage medium having stored thereon a computer program which, when executed by a processor, performs the steps of the first aspect.

The method for displaying the candidate sentence list, the device for displaying the candidate sentence list, the terminal device and the computer readable storage medium provided by the embodiment of the application have the following beneficial effects that:

in the embodiment of the application, when a user triggers a preset recall operation through a session interface, a plurality of preset recall modules are called to respectively perform candidate statement recall operations according to information input by the user on the session interface to obtain a plurality of candidate statements, feature information of each candidate statement is coded according to a preset coding strategy to obtain statement feature vectors, each candidate statement corresponds to a statement feature vector, the statement feature vectors corresponding to each candidate statement are input into a pre-trained statement sequencing model to obtain a sequencing value of each candidate statement, the sequencing value is used for indicating the high quality of the candidate statements, a candidate statement list obtained by sequencing a plurality of candidate statements according to the sequencing value of each candidate statement is displayed in the session interface, and the candidate statements are sequenced by a unified sequencing logic, the applicability of the candidate sentence list obtained after sorting is improved.

Drawings

In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed to be used in the embodiments or the prior art descriptions will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present application, and it is obvious for those skilled in the art to obtain other drawings based on these drawings without inventive exercise.

FIG. 1 is a flowchart of an implementation of a method for displaying a candidate sentence list according to an embodiment of the present application;

FIG. 2 is a flowchart of an implementation of a method for displaying a candidate sentence list according to another embodiment of the present application;

FIG. 3 is a block diagram illustrating an apparatus for displaying a candidate sentence list according to an embodiment of the present disclosure;

fig. 4 is a block diagram of a terminal device according to an embodiment of the present disclosure.

Detailed Description

In order to make the objects, technical solutions and advantages of the present application more apparent, the present application is described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the present application and are not intended to limit the present application.

In the method for displaying a candidate sentence list provided in this embodiment, the execution main body is a terminal device, and specifically, the execution main body may be a terminal device installed with an application program configured with the method function, or a terminal device displaying an operation page configured with the method function. Based on this, a target script file is configured for the application program or the operation page, and the target script file describes the method for displaying the candidate sentence list provided by this embodiment, so that the terminal device executes the target script file while running the application program or displaying the operation page, and further executes each step in the method for displaying the candidate sentence list.

During implementation, the terminal device enters a session page by running the application program or displaying an operation page, when a user triggers a preset recall operation through the session interface, the terminal device calls a plurality of preset recall modules to respectively perform candidate statement recall operations according to information input by the user on the session interface to obtain a plurality of candidate statements, the terminal device encodes feature information of each candidate statement according to a preset encoding strategy to obtain statement feature vectors, each candidate statement corresponds to a statement feature vector, the statement feature vectors corresponding to each candidate statement are input into a pre-trained statement ranking model to obtain a ranking value of each candidate statement, the ranking value is used for indicating the quality of the candidate statements, and then a candidate statement list obtained by ranking a plurality of candidate statements according to the ranking value of each candidate statement is obtained, the candidate sentences are displayed in the conversation interface, so that the candidate sentences are sorted by uniform sorting logic, and the applicability of the candidate sentence list obtained after sorting is improved.

For example, taking the above-mentioned session interface as an example of an "intelligent customer service" interface displayed on the terminal device when the user performs a session by using an "intelligent customer service" module in the application program, when the user triggers a preset recall operation through the "intelligent customer service session interface", the terminal device invokes the preset recall modules to perform a candidate sentence recall operation respectively according to information input by the user at the "intelligent customer service session interface" to obtain a plurality of candidate sentences, the terminal device encodes feature information of each candidate sentence according to a preset encoding policy to obtain a sentence feature vector, because each candidate sentence corresponds to a sentence feature vector respectively, the terminal device inputs the sentence feature vector into a pre-trained sentence ranking model, can obtain a ranking value of each candidate sentence, and the ranking value is used for indicating high or low quality of the candidate sentence, the terminal equipment sorts the candidate sentences according to the sorting value of each candidate sentence to obtain a candidate sentence list, and finally displays the candidate sentence list in an intelligent customer service session interface, so that the candidate sentences are sorted by uniform sorting logic, and the applicability of the candidate sentence list obtained after sorting is improved.

The method for displaying a candidate sentence list provided in this embodiment is described in detail below by way of specific implementation.

Fig. 1 shows a flowchart of an implementation of a method for displaying a candidate sentence list provided in an embodiment of the present application, which is detailed as follows:

s11: and responding to a preset recall operation triggered by a user through a session interface, calling a plurality of preset recall modules to respectively recall candidate sentences according to information input by the user on the session interface to obtain a plurality of candidate sentences.

In step S11, the session interface is an interface displayed on the terminal device and used for performing a session with the intelligent session system, the session interface can display all contents of the user in the session with the intelligent session system, and the session interface is further used for monitoring whether the user triggers a preset recall operation.

In all embodiments of the present application, the preset recall operation generally refers to a candidate statement recall operation triggered when a user inputs any information in the above-mentioned session interface, and if the session interface detects that the user inputs any information in the session interface, it is determined that the user has triggered the preset recall operation through the session interface, and at this time, the terminal device is made to call a plurality of preset recall modules to respectively perform the candidate statement recall operation according to the information input by the user in the session interface. Here, the plurality of recall modules are service methods for performing the recall operation of the candidate sentences according to different recall modes, that is, the specific implementation modes or implementation logics for performing the recall operation of the candidate sentences by each recall module according to the information input by the user on the session interface are different.

In practical application, a user starts the intelligent conversation system on the terminal equipment to enter a conversation interface, and after the user inputs information in the conversation interface, the intelligent conversation system can perform semantic analysis on the information input by the user so as to match reply contents related to the information, thereby realizing man-machine conversation. Here, when the session interface is displayed on the terminal device, the content input by the user is monitored by default, so that when the user inputs any information in the session interface, it can be determined that the user triggers the preset recall operation through the session interface.

It should be noted that, because the recall modes respectively followed by each of the recall modules are different, each of the recall modules performs a recall operation on the candidate sentences according to the information input by the user on the session interface, and the obtained candidate sentences also have a great difference or obvious difference.

As an example, it is assumed that the plurality of recall modules include an ES search recall module, a model search recall module and a keyword search recall module, where the ES search recall refers to a recall operation performed according to information input by a user at a session interface by using an ES algorithm, where the ES algorithm is used for describing a probability relationship between the information input by the user at the session interface and a recall result; the model searching recall describes the corresponding relationship between the information input by the user on the session interface and the recall result, the information input by the user on the session interface is taken as the input information of the model, the recall result is taken as the output of the model, and the corresponding relationship between the information input by the user on the session interface and the recall result can be changed by changing the mapping logic represented by the model; the keyword search recall is to use information input by the user in the session interface as a keyword, and use the most selected sentence by other users in the history candidate sentences as the candidate sentences, that is, the keyword search recall utilizes the selection frequency or times of the candidate sentences.

For example, the information input by the user on the session interface is the name "XYZ" of the product, an ES search recall is called to perform a candidate sentence recall operation according to the name "XYZ" of the product, and a sentence containing the name "XYZ" of the product and input by the user in the past is obtained as a candidate sentence; calling a model search recall module to perform candidate statement recall operation according to the name 'XYZ' of the product to obtain a candidate statement consisting of attribute question sentences named as 'XYZ' of the product; and calling a keyword recall module to perform candidate statement recall operation according to the name 'XYZ' of the product, and obtaining the statement which is selected by other users most and contains the name 'XYZ' as a candidate statement.

In some application scenarios, when a plurality of preset recall modules are called to respectively perform candidate statement recall operations according to information input by a user on a session interface, in order to be more suitable for a use scenario or a use habit of the user, a new recall module may be configured based on an existing recall module, or various recall modules may be configured to realize recall of candidate statements in a manner more suitable for the session scenario.

The existing multiple recall modules comprise: recall module a1, recall module a2, recall module B1, and recall module B2 are examples, and recall module a1 and recall module a2 are statistical recall modules and recall module B1 and recall module B2 are screening recall modules, and configuring a new recall module based on existing recall modules may be by nesting or combining statistical recall modules and screening recall modules.

For example, recall module B1 was nested within recall module A1 to obtain a new recall module C1, where the plurality of recall modules includes: recall module a1, recall module a2, recall module B1, recall module B2, and recall module C1. After recalling the first candidate sentence set obtained in the statistical manner according to the recall logic of the recall module a1, the new recall module C1 screens the first candidate sentence set according to the screening logic of the recall module B1 to obtain a second candidate sentence set, which is a recall result obtained by the recall module C1 performing the candidate sentence recall operation according to the information input by the user on the session interface.

It can be understood that, because there is a difference in the recall modes followed by the recall modules in the plurality of recall modules, there is inevitably a difference between the obtained plurality of candidate sentences, that is, semantic expressions or emphasis points between the candidate sentences are different, so that it is inconvenient to directly sort the plurality of candidate sentences. In order to achieve rational sorting of a plurality of candidate sentences, after a plurality of candidate sentences are obtained, encoding processing needs to be performed on each candidate sentence, and sorting operation on the plurality of candidate sentences is achieved based on the sentence feature vectors of the candidate sentences obtained through the encoding processing. As shown in fig. 1, steps S12 through S14 are performed after step S11.

S12: and coding the characteristic information of each candidate statement according to a preset coding strategy to obtain a statement characteristic vector.

In step S12, a preset encoding policy is used to describe the manner in which the feature information of the candidate sentence is encoded, and to distinguish or indicate the interfaces and tools that are called when the feature information of the candidate sentence is encoded. Here, the feature information of a candidate sentence generally refers to the feature content corresponding to each candidate sentence, and since different candidate sentences are recalled by different recall modules, the feature information of different candidate sentences in the same feature dimension also differs, so that the feature information can be used as a factor for measuring the quality of the corresponding candidate sentences.

As an example, in the process of encoding the feature information of the candidate sentences, the feature information of the candidate sentences may be clustered in advance according to feature dimensions or common attributes between the feature information, and encoded by using different encoding strategies according to the clustering result.

For example, the feature information of the candidate sentences may be clustered, and the feature information of the candidate sentences may be classified into: an immediate and/or data-based statement feature; the instant sentence features may include: sentence correlation characteristics, sentence sources and sentence content quality; the dataform statement features may include: sentence click rate and sentence content quality.

For another example, the feature information of the candidate sentence may be divided into: local sentence features and cloud sentence features; the local statement feature may include: sentence correlation characteristics, sentence sources, sentence content quality and sentence content quality; cloud statement features may include; sentence click rate, etc.

In this embodiment, the encoding is performed by using different encoding strategies according to the clustering result, specifically, the corresponding encoding strategies may be pre-configured according to the feature information, and then the encoding is performed by using different encoding strategies according to different feature information represented by the clustering result. For example, for sentence associated features, a BI-LSTM model may be used to encode sentence vector features, and sentence sources, sentence click rates, sentence content quality, and semantic features, a one-hot or word embedding may be used to obtain a vector represented by each feature, and finally all feature encoded vectors are spliced (added) to obtain a sentence feature vector of each candidate sentence.

It should be noted that the term feature vector generally refers to a vector obtained by encoding feature information corresponding to each candidate term, that is, each candidate term corresponds to a group of term feature vectors. In consideration of the fact that the scoring dimension of each candidate sentence can be unified when a plurality of candidate sentences are arranged, when feature information of each candidate sentence in the plurality of candidate sentences is coded respectively, the type or the number of coded feature information of all candidate sentences should be consistent.

In implementation, if the feature information to be encoded is multiple and different encoding modes are required, different feature information can be separately encoded when the feature information of each candidate statement is encoded, and finally all the obtained feature vectors are spliced to obtain the statement feature vector. If the feature information required to be coded is multiple, but only one coding mode is required, different feature information can be separately coded in the same coding mode when the feature information of each candidate statement is coded, and finally all the obtained feature vectors are spliced to obtain the statement feature vector.

It can be understood that, if only one kind of feature information is required to be encoded, one kind of encoding mode may be adopted to separately encode the one kind of feature information, and the obtained feature vector is used as the sentence feature vector.

As an example, in the above solution, the step S12 includes:

constructing feature information of each candidate statement to obtain a feature information set of each candidate statement;

coding the characteristic information in each characteristic information set according to a preset coding strategy to obtain a characteristic vector set;

and splicing all the feature vectors in the feature vector set to obtain the statement feature vector.

In this embodiment, the feature information set of the candidate sentence includes: sentence association characteristics, sentence source, sentence click rate, sentence content quality, and semantic characteristics. When the method is implemented, the feature information set is used as a factor for measuring the quality of the candidate sentences, and can be selected according to a specific use scene or requirement.

The feature vector set refers to all vectors obtained by encoding each feature information corresponding to each candidate sentence, that is, a feature vector set obtained by encoding the feature information set after feature extraction is performed on each candidate sentence. When the feature information of each candidate sentence in the plurality of candidate sentences is encoded respectively, the type or the number of the encoded feature information of all the candidate sentences should be consistent.

For example, the feature information of the candidate sentence may include: the sentence association feature, the sentence source, the sentence click rate, the sentence content quality and the semantic feature are assumed to be specified as feature information sets for encoding, and the sentence content quality and the semantic feature of each candidate sentence are required to be encoded when encoding the feature information in the feature information set of each candidate sentence, so that the sentence feature vectors corresponding to each candidate sentence are kept consistent in feature dimension.

In this embodiment, the feature information in each feature information set is encoded according to a preset encoding strategy, which may be to configure a corresponding encoding strategy according to the feature information, and then encode with different encoding strategies according to different feature information. For example, for sentence associated features, a BI-LSTM model may be used to encode sentence vector features, and sentence sources, sentence click rates, sentence content quality, and semantic features, a one-hot or word embedding may be used to obtain a vector represented by each feature, and finally all feature encoded vectors are spliced (added) to obtain a sentence feature vector of each candidate sentence.

In practical applications, feature information is constructed for each candidate sentence, but the content specifically included in the feature information set is specified in advance.

As an example, the set of feature information for a candidate sentence includes: sentence association characteristics, sentence source, sentence click rate, sentence content quality, and semantic characteristics.

Here, the term correlation characteristic refers to a characteristic of the conversation content before the user inputs information on the conversation interface, that is, a term correlation characteristic is obtained by extracting a history conversation term before the user currently inputs information and performing characteristic extraction on the history conversation term. During implementation, a one-hot or word Embedding coding mode is adopted, the representation of Sentence granularity and the representation of word granularity in the historical conversation sentences are fully utilized, and the sequence Embedding is obtained through Sentence granularity attention and serves as a feature vector.

The statement source refers to a recall module for recalling the candidate statement, and when the feature information architecture is carried out on the candidate statement, the recall module identification carried by the candidate statement can be identified and used as the feature content of the statement source. Here, the recall module identifier may be configured in the corresponding candidate statement when the recall module is invoked to recall the candidate statement according to information input by the user on the session interface, and the statement source is considered as a feature because different candidate statements are recalled in different manners and corresponding recall modules, so as to consider the merits and the association degrees of different recall modules in the ranking evaluation process.

The sentence click rate refers to the ratio of the number of times the candidate sentence is clicked by the user to the number of times the candidate sentence is displayed, and can reflect the probability and frequency of the candidate sentence being used to some extent. The sentence click rate can be obtained by sending a query request to the server and then receiving information returned by the server, and can also be obtained by periodically receiving updated data sent by the server, storing the updated data in a local database of the terminal equipment, and directly obtaining the updated data from the local database when feature information of candidate sentences needs to be constructed. In some embodiments, the sentence click rate may also characterize: whether the user has clicked on the candidate question, whether the user has collected the candidate question, whether the user has liked the candidate question, and the like.

Sentence content quality is used to describe how good or bad the candidate sentences are constructed, for example, sentence content quality may include: the content length of the candidate sentence, whether the h5 tag is included in the candidate sentence, the type of the candidate sentence, such as text, picture, video, etc., and whether the jump link is included in the candidate sentence, etc.

Semantic features refer to the similarity or vector distance of a candidate sentence to the information entered by the user in the conversational interface. For example: the method comprises the steps of inputting the length relation between information and a candidate sentence in a conversation interface by a user, inputting the similarity between the information and the candidate sentence in the conversation interface by the user, inputting the editing distance between the information and the candidate sentence in the conversation interface by the user, inputting the literal matching degree between the information and the candidate sentence in the conversation interface by the user, inputting the Ngram information between the information and the candidate sentence in the conversation interface by the user, and the like.

It can be understood that each candidate statement corresponds to one feature information set, the feature information in each feature information set is encoded to obtain a feature vector set, and the feature vector set necessarily corresponds to a single candidate statement, so that a unique correspondence exists between statement feature vectors obtained by splicing all feature vectors in the feature vector set and the candidate statements.

S13: inputting the sentence characteristic vector into a pre-trained sentence sequencing model to obtain a sequencing value of each candidate sentence; wherein the ranking value is used to indicate a high or low quality of each of the candidate sentences.

In step S13, the pre-trained sentence ranking model is used to determine the selected probability value of each candidate sentence, and the ranking value of each candidate sentence can be obtained according to the selected probability value of each candidate sentence.

It should be noted that the ranking value is also used to characterize the ranking position of each candidate sentence in the candidate sentence list, where the larger the probability value of the candidate sentence is selected, the smaller the corresponding ranking value is, and the more forward the position of the candidate sentence with the smaller ranking value is ranked in the candidate sentence list, the more convenient it is for the user to select.

In this embodiment, the pre-trained sentence ranking model takes the sentence feature vector as an input, and scores each candidate sentence based on the sentence feature vector. Here, because the term feature vector generally refers to a vector obtained by encoding feature information corresponding to each candidate term, that is, each candidate term corresponds to a group of term feature vectors, in the process of obtaining the ranking value of each candidate term, the term ranking model after being trained in advance takes the term feature vectors as input, and the influence degree of each feature information in the feature information set corresponding to each candidate term is not considered any more, because the term feature vector can completely represent the feature difference, the importance degree, or the applicability degree between the corresponding candidate term and other candidate terms.

In the implementation process, an existing statement feature vector model framework can be used for building a statement ordering model, wherein the existing statement feature vector model framework can comprise a vector input layer, a coding layer, a decoding layer and a full connection layer, in order to enable the built statement ordering model to take the ordering value of each candidate statement as output, a separator layer is further added in the existing statement feature vector model framework and used for obtaining the ordering value of each candidate statement according to the selected probability value of each candidate statement output by the full connection layer. After the sentence sequencing model is built, training the sentence sequencing model by using a pre-built training sample, wherein the training sample comprises the characteristic sample of the sample sentence, and the artificially set probability value and the sample sequencing value of the characteristic sample of each sample sentence under different proportion combinations.

As an example, in the above solution, the step S13 includes:

obtaining a selected probability value of each candidate statement according to the statement feature vector;

and obtaining the ranking value of each candidate sentence based on the selected probability value of each candidate sentence.

In this embodiment, the selected probability value is used to represent the probability that the candidate sentence is possibly selected by the user, where the larger the selected probability value is, the smaller the corresponding ranking value is, and the more forward the position of the candidate sentence with the smaller ranking value in the candidate sentence list is, the more likely it is that the candidate sentence is selected by the user.

In implementation, the sorting value of each candidate sentence is obtained based on the selected probability value of each candidate sentence, which may be sorting the candidate sentences according to the selected probability value of each candidate sentence, where in order to obtain the sorting value of each candidate sentence more quickly, the candidate sentences are sorted from large to small based on the selected probability value of each candidate sentence, and the sorting value of each candidate sentence is determined according to the sorting result.

It should be understood that, in practical application, a plurality of candidate sentences may also be sorted from small to large based on the selected probability value of each candidate sentence to obtain a sorting result, a final sorting result is obtained by filtering the candidate sentences of the first several bits of the sorting result, the candidate sentences with smaller selected probability values can be filtered, and then the sorting value is obtained by the reverse ranking of the final sorting result.

S14: and displaying a candidate sentence list obtained by sequencing the candidate sentences according to the sequencing value of each candidate sentence in the conversation interface.

In step S14, the candidate sentence list is obtained by sorting the plurality of candidate sentences based on the sorting value corresponding to each candidate sentence. Here, all candidate sentences presented in the candidate sentence list may be part or all of the plurality of candidate sentences.

In this embodiment, when the candidate sentence list is displayed in the conversation interface, the display manner of the candidate sentence list may be adjusted according to the actual number of the plurality of candidate sentences. For example, by creating a window in which a candidate sentence list is displayed, the candidate sentence list is directly loaded and displayed in the window. For another example, by creating a window in which a candidate sentence list is displayed, loading the candidate sentence list in the window, only the contents of a part of the candidate sentence list are displayed.

It can be understood that, in practical applications, different sorting strategies may also be configured according to the type of the feature information specified in the sentence feature set, that is, the influence of different feature information on sorting is considered. For example, when the sentence feature vector only contains a corresponding vector for describing the content quality feature of the candidate sentence, the ranking value may also be used to indicate the quality of the candidate sentence. When the rank value is used to indicate the quality of a candidate sentence, the smaller the rank value, the higher the quality of the candidate sentence.

As an embodiment of the present application, step S14 includes:

sorting the candidate sentences according to the sorting value of each candidate sentence to obtain a candidate sentence list;

and displaying part of or all of the contents of the candidate sentence list in the conversation interface according to a preset display number.

In this embodiment, the candidate sentence list is used to regularly display a plurality of candidate sentences, and the display position of each candidate sentence in the candidate sentence list corresponds to the ranking value thereof. The preset display number is used for describing the number of candidate sentences which can be displayed after the candidate sentence list is refreshed once.

It can be understood that, in practical applications, the preset number of presentations may be set according to the size of the session interface and/or the length of the candidate sentences. For example, when the session interface is small, the preset number of presentations may be set to a small value. For another example, when the session interface is large and the content of the candidate sentence is short, the preset number of presentations may be set to a large value.

As a possible implementation manner of this embodiment, the foregoing steps: displaying part or all of the contents of the candidate sentence list in the conversation interface according to a preset display number, including:

when the number of the candidate sentences in the candidate sentence list is larger than a preset display number, displaying partial contents of the candidate sentence list in the conversation interface;

and when the number of the candidate sentences in the candidate sentence list is equal to or less than the preset display number, displaying partial content of the candidate sentence list in the conversation interface.

In this embodiment, when the number of candidate sentences that need to be displayed in the candidate sentence list is greater than the preset display number, part of the content of the candidate sentence list is displayed in the conversation interface, and another part of the candidate sentences that are not displayed may not be loaded or displayed temporarily, and when a user browses the candidate sentence list, the user may browse the candidate sentences that are not displayed initially by dragging the content in the candidate sentence list. Here, the preset display number is configured to limit the number of candidate sentences displayed in the candidate sentence list, so that excessive candidate sentence contents can be prevented from being loaded at one time when the number of candidate sentences is large, excessive screen contents can be prevented from being shielded, and excessive system resources do not need to be occupied.

In the above scheme, when a user triggers a preset recall operation through a session interface, a preset plurality of recall modules are called to respectively perform candidate sentence recall operations according to information input by the user on the session interface to obtain a plurality of candidate sentences, feature information of each candidate sentence is encoded according to a preset encoding strategy to obtain sentence feature vectors, because each candidate sentence corresponds to a sentence feature vector, the sentence feature vectors corresponding to each candidate sentence are input into a pre-trained sentence sequencing model to obtain a sequencing value of each candidate sentence, the sequencing value is used for indicating the high quality of the candidate sentences, and then a candidate sentence list obtained by sequencing the candidate sentences according to the sequencing value of each candidate sentence is displayed in the session interface to realize the sequencing of the candidate sentences by a uniform sequencing logic, the applicability of the candidate sentence list obtained after sorting is improved.

Fig. 2 shows a flowchart of an implementation of a method for displaying a candidate sentence list according to another embodiment of the present application. Based on any of the above embodiments, the present embodiment further includes steps S21 to S22 before step S11. The details are as follows:

s21: configuring a pre-configured recall strategy script file into an initial session interface file to obtain a session interface file; the recall strategy script file is used for describing an execution environment and an execution process for calling the plurality of preset recall modules to perform candidate statement recall operation;

s22: and if the session interface file is loaded and the session interface is displayed, executing the recall strategy script file.

In this embodiment, the initial session interface file is used to load and display the session interface on the terminal device. The initial session interface file may also generally refer to a session interface file that is not configured with a corresponding recall function. The recall strategy script file is used for describing an execution environment and an execution process for invoking a plurality of preset recall modules to perform candidate statement recall operation, so that after the recall strategy script file is configured to the initial session interface file, the obtained session interface file can be loaded and displayed on a session interface based on the content of the initial session interface file, the recall strategy script is executed, the execution environment is further built according to the content of the initial session interface file, and the plurality of preset recall modules are invoked to perform candidate statement recall operation according to the corresponding execution process.

In practical application, the pre-configured recall policy script file is sent to the terminal device by the server, when a user requests to perform a session through the terminal device, the pre-configured recall policy script file is sent to the terminal device by the server, and the terminal device configures the pre-configured recall policy script file into an initial session interface file to obtain a session interface file. Here, the preconfigured recall policy script file is preconfigured in the server, and after the server sends the preconfigured recall policy script file to the terminal device, and after the developer updates the preconfigured recall policy script file, the server may also send the preconfigured recall policy script file to the terminal device again when the terminal device requests to access the session page, that is, send the updated preconfigured recall policy script file to the terminal device, so that the terminal device can configure the updated preconfigured recall policy script file into the initial session interface file to obtain a new session interface file.

For example, when a user accesses a server through a terminal device and requests a session activity, the server sends the pre-configured recall policy script file to the terminal device, the terminal device configures the recall policy script file to an initial session interface file after receiving the recall policy script file, so as to obtain a session interface file, and the terminal device loads the session interface file, so as to display a session interface and execute the recall policy script file.

In practical applications, the developer may pre-configure the recall policy script file on the server side according to the recall requirement, for example, configure different recall policies, recall channels, recall ranges, etc. of the recall policy script file according to the recall requirement.

It can be understood that, when configuring the recall policy script file, a person skilled in the art needs to consider the compatibility problem between the recall policy script file and the initial session interface file, so that only the terminal device needs to be ensured to normally execute the recall policy script file when loading the session interface file, and therefore how to configure the recall policy script file is not described herein again.

In addition, the pre-configured recall strategy script file is configured into the initial session interface file to obtain the session interface file, and the recall strategy script file is used for describing an execution environment and an execution process for calling a plurality of preset recall modules to perform candidate statement recall operation, so that the recall operation of the candidate statements in different modes and different channels can be realized by executing the recall strategy script file when the session interface is loaded and displayed on the terminal equipment, the efficiency and the scientific degree of the recall of the candidate statements are improved, and omission of the recall of the candidate statements is avoided.

Referring to fig. 3, fig. 3 is a block diagram illustrating a device for displaying a candidate sentence list according to an embodiment of the present application. The mobile terminal in this embodiment includes units for executing the steps in the embodiments corresponding to fig. 1 and fig. 2. Please refer to fig. 1 and fig. 2, and fig. 1 and fig. 2 for the corresponding embodiments. For convenience of explanation, only the portions related to the present embodiment are shown. Referring to fig. 3, the apparatus 30 for displaying a candidate sentence list includes: recall unit 31, encoding unit 32, sorting unit 33, and display unit 34. Specifically, the method comprises the following steps:

the recall unit 31 is configured to, in response to a preset recall operation triggered by a user through a session interface, invoke a plurality of preset recall modules to respectively perform a candidate statement recall operation according to information input by the user on the session interface, so as to obtain a plurality of candidate statements;

the encoding unit 32 is configured to encode the feature information of each candidate statement according to a preset encoding strategy to obtain a statement feature vector;

the sorting unit 33 is configured to input the sentence feature vector into a pre-trained sentence sorting model to obtain a sorting value of each candidate sentence; wherein the ranking value is used to indicate a high or low quality of each of the candidate sentences;

a display unit 34, configured to display a candidate sentence list obtained by sorting the multiple candidate sentences according to the sorting value of each candidate sentence in the conversation interface.

As an embodiment, the apparatus 30 for displaying a candidate sentence list further includes: a configuration unit 35 and an execution unit 36. Specifically, the method comprises the following steps:

the configuration unit 35 is configured to configure a pre-configured recall policy script file into an initial session interface file to obtain a session interface file; the recall strategy script file is used for describing an execution environment and an execution process for calling the plurality of preset recall modules to perform candidate statement recall operation;

and the executing unit 36 is configured to execute the recall policy script file if the session interface file is loaded and the session interface is displayed.

It should be understood that, in the structural block diagram of the apparatus for displaying a candidate sentence list shown in fig. 3, each unit is used to execute each step in the embodiment corresponding to fig. 1 and 2, and for each step in the embodiment corresponding to fig. 1 and 2, the above embodiment has been explained in detail, specifically please refer to the description in the embodiments corresponding to fig. 1 and 2 and fig. 1 and 2, and details are not repeated here.

Fig. 4 is a block diagram of a terminal device according to an embodiment of the present disclosure. As shown in fig. 4, the terminal device 40 of this embodiment includes: a processor 41, a memory 42 and a computer program 43 stored in said memory 42 and executable on said processor 41, such as a program of a method of displaying a list of candidate sentences. The steps in each embodiment of the method for displaying a candidate sentence list, such as S11 to S14 shown in fig. 1, and further such as S21, S22, and S11 to S14 shown in fig. 2, are implemented when the processor 41 executes the computer program 43, and the functions of each unit in the embodiment corresponding to fig. 3, such as the functions of the units 31 to 36 shown in fig. 3, are implemented when the processor 41 executes the computer program 43, for example, please refer to the relevant description in the embodiment corresponding to fig. 3, which is not repeated herein.

Illustratively, the computer program 43 may be divided into one or more units, which are stored in the memory 42 and executed by the processor 41 to accomplish the present application. The one or more units may be a series of computer program instruction segments capable of performing specific functions, which are used to describe the execution process of the computer program 43 in the terminal device 40. For example, the computer program 43 may be divided into a recall unit, an encoding unit, a sorting unit, a display unit, a configuration unit, and an execution unit, and the specific functions are as described above.

The turntable device may include, but is not limited to, a processor 41, a memory 42. Those skilled in the art will appreciate that fig. 4 is merely an example of a terminal device 40 and does not constitute a limitation of terminal device 40 and may include more or fewer components than shown, or some components in combination, or different components, e.g., the turntable device may also include input output devices, network access devices, buses, etc.

The Processor 41 may be a Central Processing Unit (CPU), other general purpose Processor, a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), an off-the-shelf Programmable Gate Array (FPGA) or other Programmable logic device, discrete Gate or transistor logic, discrete hardware components, etc. A general purpose processor may be a microprocessor or the processor may be any conventional processor or the like.

The storage 42 may be an internal storage unit of the terminal device 40, such as a hard disk or a memory of the terminal device 40. The memory 42 may also be an external storage device of the terminal device 40, such as a plug-in hard disk, a Smart Media Card (SMC), a Secure Digital (SD) Card, a Flash memory Card (Flash Card), and the like, which are provided on the terminal device 40. Further, the memory 42 may also include both an internal storage unit and an external storage device of the terminal device 40. The memory 42 is used for storing the computer program and other programs and data required by the turntable device. The memory 42 may also be used to temporarily store data that has been output or is to be output.

The above-mentioned embodiments are only used for illustrating the technical solutions of the present application, and not for limiting the same; although the present application has been described in detail with reference to the foregoing embodiments, it should 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; such modifications and substitutions do not substantially depart from the spirit and scope of the embodiments of the present application and are intended to be included within the scope of the present application.

Claims

1. A method of displaying a list of candidate sentences comprising:

2. The method according to claim 1, wherein before the step of invoking a plurality of preset recall modules to respectively perform a candidate sentence recall operation according to information input by a user on the session interface in response to a preset recall operation triggered by the user through the session interface to obtain a plurality of candidate sentences, the method further comprises:

configuring a pre-configured recall strategy script file into an initial session interface file to obtain a session interface file; the recall strategy script file is used for describing an execution environment and an execution process for calling the plurality of preset recall modules to perform candidate statement recall operation;

and if the session interface file is loaded and the session interface is displayed, executing the recall strategy script file.

3. The method of claim 1, wherein the encoding the feature information of each candidate sentence according to a preset encoding policy to obtain a sentence feature vector comprises:

4. The method of claim 1, wherein the inputting the sentence feature vector into a pre-trained sentence ordering model to obtain an ordering value of each candidate sentence comprises:

5. The method according to claim 1, wherein the step of displaying the candidate sentence list obtained by sorting the plurality of candidate sentences according to the sorting value of each candidate sentence in the conversation interface comprises:

6. The method of claim 5, wherein the displaying a part of or all of the contents of the candidate sentence list in the conversation interface according to a preset number of presentations comprises:

7. The method of displaying a list of candidate sentences according to any one of claims 1 to 6, wherein the set of feature information of the candidate sentences includes: sentence association characteristics, sentence source, sentence click rate, sentence content quality, and semantic characteristics.

8. An apparatus for displaying a list of candidate sentences comprising:

9. A terminal device, characterized in that the terminal device comprises a memory, a processor and a computer program stored in the memory and executable on the processor, the processor executing the computer program with the steps of the method according to any of claims 1 to 7.

10. A computer-readable storage medium, in which a computer program is stored which, when being executed by a processor, carries out the steps of the method according to any one of claims 1 to 7.