WO2023155678A1 - 用于确定信息的方法和装置 - Google Patents
用于确定信息的方法和装置 Download PDFInfo
- Publication number
- WO2023155678A1 WO2023155678A1 PCT/CN2023/073922 CN2023073922W WO2023155678A1 WO 2023155678 A1 WO2023155678 A1 WO 2023155678A1 CN 2023073922 W CN2023073922 W CN 2023073922W WO 2023155678 A1 WO2023155678 A1 WO 2023155678A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- information
- candidate
- consistency
- candidate answer
- reply
- Prior art date
Links
- 238000000034 method Methods 0.000 title claims abstract description 57
- 239000013598 vector Substances 0.000 claims description 80
- 238000003860 storage Methods 0.000 claims description 16
- 230000015654 memory Effects 0.000 claims description 15
- 238000004590 computer program Methods 0.000 claims description 9
- 230000004044 response Effects 0.000 claims description 6
- 238000012549 training Methods 0.000 description 11
- 230000006870 function Effects 0.000 description 10
- 235000013305 food Nutrition 0.000 description 9
- 230000007246 mechanism Effects 0.000 description 8
- 238000004891 communication Methods 0.000 description 6
- 238000010586 diagram Methods 0.000 description 5
- 239000011159 matrix material Substances 0.000 description 5
- 230000008447 perception Effects 0.000 description 4
- 230000008569 process Effects 0.000 description 4
- 238000004458 analytical method Methods 0.000 description 3
- 238000012986 modification Methods 0.000 description 3
- 230000004048 modification Effects 0.000 description 3
- 241000699666 Mus <mouse, genus> Species 0.000 description 2
- 238000013473 artificial intelligence Methods 0.000 description 2
- 230000002457 bidirectional effect Effects 0.000 description 2
- 230000009194 climbing Effects 0.000 description 2
- 230000008094 contradictory effect Effects 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 230000003993 interaction Effects 0.000 description 2
- 239000004973 liquid crystal related substance Substances 0.000 description 2
- 230000014759 maintenance of location Effects 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- 238000012546 transfer Methods 0.000 description 2
- 241000699670 Mus sp. Species 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 230000001413 cellular effect Effects 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000010411 cooking Methods 0.000 description 1
- 238000013500 data storage Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000003203 everyday effect Effects 0.000 description 1
- 239000000835 fiber Substances 0.000 description 1
- 238000010295 mobile communication Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000011176 pooling Methods 0.000 description 1
- 230000001953 sensory effect Effects 0.000 description 1
- 235000021259 spicy food Nutrition 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 230000017105 transposition Effects 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/33—Querying
- G06F16/332—Query formulation
- G06F16/3329—Natural language query formulation or dialogue systems
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/33—Querying
- G06F16/3331—Query processing
- G06F16/334—Query execution
- G06F16/3344—Query execution using natural language analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/30—Semantic analysis
Definitions
- the present disclosure relates to the field of computer technology, in particular to a method and device for determining information.
- the present disclosure provides a method, an apparatus, an electronic device, a computer-readable storage medium, and a computer program product for determining information.
- Some embodiments of the present disclosure provide a method for determining information, including: acquiring overall context information and a plurality of candidate answer information for replying to the overall context information; for each candidate in the plurality of candidate answer information Answer information, determine the total difference information between the candidate answer information and other candidate answer information except the candidate answer information in multiple candidate answer information; the consistency information between the total difference information and the overall above information , determined as the first consistency information between the candidate answer information and the overall above information; according to the first consistency information corresponding to each candidate answer information, determine the target answer information from multiple candidate answer information.
- Some embodiments of the present disclosure provide an apparatus for determining information, including: obtaining a single A unit configured to acquire the overall context information and a plurality of candidate answer information for replying to the overall context information; a first determining unit configured to determine the candidate answer information for each candidate answer information in the plurality of candidate answer information Total difference information between the reply information and other candidate reply information except the candidate reply information among the plurality of candidate reply information; the second determining unit is configured to combine the total difference information with the overall above information Consistency information is determined as the first consistency information between the candidate answer information and the overall above information; the third determining unit is configured to, according to the first consistency information corresponding to each candidate answer information, from multiple The target answer information is determined in the candidate answer information.
- Some embodiments of the present disclosure provide an electronic device, including: one or more processors: a storage device for storing one or more programs, when the one or more programs are executed by the one or more processors, so that One or more processors implement the method for determining information as provided in the first aspect.
- Some embodiments of the present disclosure provide a computer-readable storage medium on which a computer program is stored, wherein, when the program is executed by a processor, any embodiment of the method for determining information as described above is implemented.
- Some embodiments of the present disclosure provide a computer program product comprising a computer program which, when executed by a processor, implements any of the embodiments of the method for determining information as described above.
- FIG. 1 is an exemplary system architecture diagram to which an embodiment of the present application can be applied;
- FIG. 2 is a flowchart of one embodiment of a method for determining information according to the present application
- FIG. 3 is a flowchart of another embodiment of a method for determining information according to the present application.
- FIG. 4 is a flowchart of an application scenario of a method for determining information according to the present application
- FIG. 5 is a schematic diagram of a training device for a model used in an application scenario in a method for determining information according to the present application;
- Fig. 6 is a schematic structural diagram of an embodiment of a device for determining information according to the present application.
- Fig. 7 is a block diagram of an electronic device used to implement the method for determining information according to the embodiment of the present application.
- FIG. 1 shows an exemplary system architecture 100 to which embodiments of the method for determining information or the apparatus for determining information of the present application can be applied.
- a system architecture 100 may include terminal devices 101 , 102 , 103 , a network 104 and a server 105 .
- the network 104 is used as a medium for providing communication links between the terminal devices 101 , 102 , 103 and the server 105 .
- Network 104 may include various connection types, such as wires, wireless communication links, or fiber optic cables, among others.
- terminal devices 101 , 102 , 103 Users can use terminal devices 101 , 102 , 103 to interact with server 105 via network 104 to receive or send messages and the like.
- the terminal devices 101, 102, and 103 may be user terminal devices on which various client applications, such as image applications, video applications, shopping applications, chat applications, search applications, and financial applications, can be installed.
- Terminal devices 101, 102, 103 may be various electronic devices with display screens and support for receiving server messages, including but not limited to smart phones, tablet computers, e-book readers, electronic players, laptop computers and desktop computers etc.
- the terminal devices 101, 102, and 103 may be hardware or software. When the terminal devices 101, 102, 103 are hardware, they may be various electronic devices, and when the terminal devices 101, 102, 103 are software, they may be installed in the electronic devices listed above. It may be implemented as multiple software or software modules (for example, multiple software modules for providing distributed services), or as a single software or software module. No specific limitation is made here.
- the server 105 can obtain the overall context information and multiple After the candidate reply information, for each candidate reply information in the plurality of candidate reply information, determine the total difference information between the candidate reply information and multiple candidate reply information, and determine the total difference information between the total difference information and the overall above information
- the consistency information is determined as the first consistency information between the candidate answer information and the overall above information
- the target answer is determined from multiple candidate answer information according to the first consistency information corresponding to each candidate answer information information.
- the method for determining information provided by the embodiments of the present disclosure may be executed by the server 105 , and correspondingly, the device for determining information may be set in the server 105 .
- terminal devices, networks and servers in Fig. 1 are only illustrative. According to the implementation needs, there can be any number of terminal devices, networks and servers.
- a flow 200 of an embodiment of a method for determining information according to the present disclosure is shown, including the following steps:
- Step 201 acquire the overall context information and multiple candidate answer information for replying the overall context information.
- the executing subject of the method for determining information may obtain the overall context information and multiple candidate answer information for replying to the overall context information.
- the overall above information refers to the historical information in the dialogue, such as historical question and answer messages in the intelligent customer service system, historical dialogue records in the intelligent chat system, etc.
- the reply information refers to the reply information used to reply the last piece of information in the whole above information, or all the information.
- Step 202 for each candidate reply information in the plurality of candidate reply information, determine total difference information between the candidate reply information and other candidate reply information in the plurality of candidate reply information except the candidate reply information.
- the relationship between the piece of candidate reply information and other candidate reply information in the plurality of candidate reply information except the candidate reply information may be determined.
- Total difference information For example, if the candidate answer information A is "the weather is sunny today", the candidate answer information B is “the weather is sunny today, the humidity is moderate”, the candidate answer information C is “the weather is sunny today, suitable for travel”, and the candidate answer information D is "today's weather Sunny, suitable for travel, suitable for mountain climbing", then the total difference information between the candidate answer information D and other candidate answer information except the candidate answer information in multiple candidate answer information is "suitable for travel, suitable for mountain climbing".
- Step 203 determine the consistency information between the total difference information and the overall above information as First consistency information between the candidate answer information and the overall above information.
- the consistency information between the candidate answer information and the overall above information may be determined based on the total difference information between the candidate answer information and other candidate answer information except the candidate answer information,
- the consistency information can be called the first consistency information, which is used to characterize the logical consistency between the content expressed by the total difference information and the content expressed by the overall above information /degree of similarity, or the degree of logical consistency/similarity between the grammar/semantics of the total difference information and the grammar/semantics of the overall above information, etc.
- Step 204 Determine target reply information from multiple candidate reply information according to the first consistency information corresponding to each candidate reply information.
- the target reply information may be determined from multiple candidate reply information according to the first consistency information corresponding to each candidate reply information, wherein the first consistency information corresponding to any reply information is Refers to: the total difference information between the arbitrary candidate answer information and other candidate answer information, and the consistency information between the overall above information.
- the method for determining information provided in this embodiment is to obtain the overall context information and a plurality of candidate answer information used to reply to the overall context information; for each candidate answer information in the plurality of candidate answer information, determine the candidate answer information, and the total difference information between other candidate answer information except the candidate answer information in multiple candidate answer information; the consistency information between the total difference information and the overall above information is determined as the candidate answer information
- the first consistency information with the overall above information according to the first consistency information corresponding to each candidate answer information, the target answer information is determined from multiple candidate answer information, and the determined target answer information can be enhanced Consistency in logic and content with the overall above information in the historical dialogue records, thereby improving the accuracy of push information.
- each candidate reply information in the plurality of candidate reply information determine total difference information between the candidate reply information and other candidate reply information in the plurality of candidate reply information except the candidate reply information, Including: obtaining the word vector of each candidate answer information; for each candidate answer information, based on the similarity between the word vector of the candidate answer information and the word vector of each other candidate answer information in other candidate answer information, Determine the difference information between the candidate answer information and each other candidate answer information; determine the difference information between the candidate answer information and all other answer information as the candidate answer information and other candidate answer information The total difference information between complex information.
- the word vector of each candidate answer information can be obtained based on a pre-trained semantic analysis model or a pre-trained language representation model, so that for each candidate answer information, based on the word vector of the candidate answer information
- the similarity between the vector and the word vectors of each other candidate answer information in multiple candidate answers determines the difference information between the candidate answer information and each candidate answer information
- the difference information can be the information of the two
- the difference content between the expressed semantics, or the difference information between the logic expressed by the two information, after that, the difference information between the candidate answer information and each other candidate answer information is summarized/collected to obtain the candidate
- the total difference information between the reply information and all other reply information that is, the total difference information between the candidate reply information and other candidate reply information except the candidate reply information among the plurality of candidate reply information.
- the difference information between the candidate reply information is determined based on the similarity between the word vectors of the candidate reply information, which can improve the accuracy and efficiency of determining the difference information.
- the consistency information between the total difference information and the overall above information is determined as the first consistency information between the candidate answer information and the overall above information, including: obtaining the word of the overall above information Vector: determine the consistency information between the total difference information and the word vector of the overall above information as the first consistency information between the candidate answer information and the overall above information.
- the word vector of the overall above information can be obtained based on the pre-trained semantic analysis model, or the pre-trained language representation model, etc., and then based on the word vector of the overall above information, and the candidate answer information and The total difference information between other candidate answer information determines the first consistency information between the candidate answer information and the overall above information. It can be understood that since the total difference information is determined based on the difference between the word vectors of the candidate answer information, the data format of the total difference information is in the form of word vectors, and the word vector based on the overall above information can be directly consistent with the total difference information Sex comparison, that is, similarity/identity comparison, to obtain the consistency information between the candidate information and the overall above information.
- the first consistency information between the candidate answer information and the overall context information is determined based on the word vector of the overall context information and the total difference information, which can improve the accuracy and efficiency of determining the first consistency information.
- FIG. 3 shows a flow 300 of another embodiment of the method for determining information according to the present disclosure, including the following steps:
- Step 301 acquiring the overall context information and a plurality of candidate answer information for replying the overall context information.
- Step 302 for each candidate reply information in the plurality of candidate reply information, determine total difference information between the candidate reply information and other candidate reply information in the plurality of candidate reply information except the candidate reply information.
- Step 303 Determine the consistency information between the total difference information and the overall above information as the first consistency information between the candidate reply information and the overall above information.
- step 301, step 302, and step 303 in this embodiment is consistent with the description of step 201, step 202, and step 203, and will not be repeated here.
- Step 304 Determine the consistency information between the total difference information and the target context information as the second consistency information between the candidate reply information and the target context information, wherein the target context information is the overall context information Information in the message that belongs to the same interlocutor as the candidate reply message.
- the consistency information between the total difference information and the target context information can be determined as the consistency information between the candidate reply information and the target context information, and for the convenience of distinction, the consistency information can be It is called the second consistency information, which is used to characterize the logical consistency between the content expressed by the total difference information and the content expressed by the target above information, or the syntax/semantics, Logical consistency with the syntax/semantics of the target above information, etc.
- the target above information refers to the dialog records generated by the interlocutor of the candidate answer information in the overall above information, for example, the overall above information is A , B dialogue between the two (A can be a real user, used to ask questions, B can be a robot customer service, used to reply to the user's question), if the current dialogue process is in the stage where B needs to answer, that is, the current If B needs to generate a message, the interlocutor of the candidate reply information is B. Therefore, the target context information is all the messages from B in the overall context information.
- Step 305 Determine target answer information from multiple candidate answer information according to the first consistency information and second consistency information corresponding to each candidate answer information.
- the target reply information may be determined from multiple candidate reply information according to the first consistency information and the second consistency information corresponding to each candidate reply information, wherein the target reply information corresponding to any reply information
- the first consistency information refers to: the total difference information between the arbitrary candidate answer information and other candidate answer information, and the consistency information between the overall above information; the second consistency information corresponding to any answer information means: the arbitrary candidate answer information and other candidate The total difference information between the selected answer information and the consistency information between the above information and the target.
- the method for determining information provided by this embodiment compared with the method described in the embodiment of FIG.
- the target reply information is also based on the second consistency information, which can enhance the logical and content consistency between the determined candidate reply information and the target above information in the historical dialogue record, that is, enhance the candidate reply information provided by the interlocutor itself. Consistency of information, thereby improving the accuracy of determining the response information.
- the total difference information is determined based on the similarity between the word vector of the candidate answer information and the word vectors of other candidate answer information except the candidate answer information in the plurality of candidate answer information, and the total difference information and
- the consistency information between the above information of the target is determined as the second consistency information between the candidate reply information and the above information of the target, including: obtaining the word vector of the above information of the target; combining the total difference information with the above information of the target.
- the consistency information between the word vectors of the text information is determined as the second consistency information between the candidate answer information and the target context information.
- the word vector of the target context information can be obtained based on a pre-trained semantic analysis model, or a pre-trained language representation model, etc., and then based on the word vector of the target context information, and the candidate answer information and The total difference information between other candidate answer information determines the second consistency information between the candidate answer information and the target above information. It can be understood that since the total difference information is determined based on the difference between the word vector of the candidate answer information and the word vector of other candidate answer information, the data format of the total difference information is in the form of word vector, based on the target context information The word vector and the total difference information can be directly compared for consistency, that is, similarity/identity comparison, to obtain the consistency information between the candidate information and the target above information.
- the second consistency information between the candidate answer information and the overall context information is determined based on the word vector of the target context information and the total difference information, which can improve the accuracy and efficiency of determining the second consistency information.
- determining the target answer information from multiple candidate answer information includes: for each candidate answer information, using the candidate The first consistency information corresponding to the answer information, the second consistency information corresponding to the candidate answer information, and the third consistency information between the semantics of the candidate answer information and the semantics of the overall above information, determine the Scores of candidate answer information; according to the score of each candidate answer information, target answer information is determined from multiple candidate answer information.
- the first consistency information corresponding to the candidate answer information, the second consistency information corresponding to the candidate answer information, and the semantic and overall information of the candidate answer information can be used.
- the semantics of the answer information is proportional to the degree of relevance/consistency between the semantics of the overall above information), and is proportional to the first consistency information (the higher the degree of first consistency, the candidate answer information The higher the score), it is proportional to the second consistency information (the higher the second consistency degree, the higher the score of the candidate answer information), after determining the score of each candidate answer information, the candidate with the highest score will be
- the reply information is determined as target reply information.
- the above candidate answer information with greater semantic relevance, higher first degree of consistency, and higher second degree of consistency is determined as the target answer information, which can improve the accuracy of determining the target answer information .
- methods for determining information include:
- the first step is to input the historical dialogue/overall information and multiple candidate reply information used to answer the overall above information into the BERT (Bidirectional Encoder Representations from Transformer, bidirectional encoding representation based on the converter) model to obtain The word vector of the overall above information output by the BERT model, and the word vector of each candidate answer information among multiple candidate answer information.
- BERT Bidirectional Encoder Representations from Transformer, bidirectional encoding representation based on the converter
- It specifically includes: obtaining (U; r i ), where U ⁇ u 1 ,..., u N ⁇ represents the overall above information, Represents the i-th sentence of the above sentence, Represents the Nth word in the overall above information of the i-th sentence, Represents the i-th candidate reply information, Represents the Nth word in the candidate answer information of the i-th sentence, after connecting the overall above information U and each candidate answer information r i , input it into the pre-trained BERT model, and obtain the overall above information output by the BERT model Word vectors and word vectors of candidate answer information: where BERT( ) returns the output of the last layer of the BERT model.
- H U represents the word vector of the overall above information U
- the total vector of the semantic information of the overall above information can also be obtained
- a fine-grained reply comparison is performed based on the two.
- ⁇ represents element-wise multiplication between two matrices.
- ⁇ represents element-wise multiplication between two matrices.
- the contrast information between r i and r j is defined as: in, Indicates the part that represents the similarity between r i and r j , Indicates the difference between r i and r j , That is, the difference information between r i and r j .
- the above difference information is integrated to obtain the differential representation of r i and all other candidate reply information That is the total difference information: in, Is the intermediate variable of the calculation formula; [ ⁇ ] represents the splicing operation, that is, to splice the sequence in []; ⁇ ( ⁇ ) represents the Dirac delta function; W 2 , W 3 , b 2 represent the model parameters.
- the third step is to carry out the consistent reasoning of the overall above logic: based on the differential representation of r i and all other candidate answer information Determine the consistency information between the candidate answer information r i and the overall above information (specifically, the word vector H U ) (ie, the first consistency information referred to above).
- H i_h Relu(A i_h H U W 7 )
- W 4 , W 5 , W 6 , W 7 , W 8 , W 9 , b 4 are model parameters;
- a h_i and A i_h are the word-level attention between all the overall above information U and candidate answer information r i Force matrix, they focus on different perspectives, that is, the attention model used to obtain the two data focuses on different information, and the attention model used to obtain A h_i focuses on the overall context related to the reply information Information
- the attention model used to obtain A i_h focuses on the reply information related to the overall above information.
- MaxPooling represents the maximum pooling operation
- SoftMax( ⁇ ) represents a normalized exponential function
- Relu( ⁇ ) represents a linear rectification function
- H h_i is the above representation of the reply perception
- H i_h is the reply representation of the above perception
- g hi represents the gate mechanism value obtained by fusing H h_i and H i_h based on the gate mechanism.
- the overall information consistency of the interlocutor is enhanced: based on the differential representation of r i and all other candidate reply information Determine the consistency information between the candidate answer information r i and the target context information (specifically, the word vector H S ) (that is, the second consistency information referred to above), wherein, the target context information refers to the overall context information given by the interlocutor to whom the candidate answer information belongs (for example, the candidate answer information is user A's need to perform dialogue information, the target above information refers to the information sent by all user A in the overall above information)
- H i_s Relu(A i_h H u W 13 )
- a s_i and A i_s are word-level attention matrices between the target above information (that is, the overall above information of the interlocutor) and the candidate reply information r i , and they focus on different perspectives, that is, to obtain two
- the attention model used for these two kinds of data focuses on different information.
- the attention model used to obtain A s_i focuses on the information above the target
- the attention model used to obtain A i_s focuses on the information related to the above information of the target.
- the reply information (reply information); the mark T represents the transposition of the matrix; H s_i represents the speaker representation of the reply perception, which is the vector representation of the overall above information, and this vector representation is related to the reply information; H i_s represents the speaker perception
- the reply representation is a vector representation of the reply, and this vector is related to the above information of the target; W 10 , W 11 , W 12 , and W 13 are model parameters.
- g si a(E s_i W 14 +E i_s W 15 +b 5 ) Among them, W 14 , W 15 , b 5 , and b 6 are model parameters; g si represents the gate mechanism value obtained by fusing H s_i and H i_s based on the gate mechanism.
- the fifth step first, the total vector of the semantics of the overall above information Consistency information between the candidate answer information r i and the overall above information Consistency information between the candidate reply information r i and the target previous information
- the three data are spliced to obtain the reasoning information H i of the candidate answer information r i :
- U, R) of the candidate answer information r i is predicted as
- W 16 and b 6 are model parameters
- the set M is the identification of candidate reply information except the candidate reply information ri among multiple candidate reply information
- R represents the set of candidate reply information
- the above-mentioned model parameters can be obtained based on model training, and the system for training the model can be shown in Figure 5.
- the steps of training the model include: obtaining sample dialogue data, including dialogue history and candidate replies, and analyzing the sample dialogue The data is filtered to filter out incomplete data and redundant data to obtain training data; the training data is input into the knowledge transfer predictor to obtain the word vector of each piece of training data, and the training data processed by the knowledge transfer predictor is input into The knowledge-aware generator obtains knowledge-aware information (such as the above-mentioned first consistency information and second consistency information), and determines target answer information based on the knowledge-aware information.
- knowledge-aware information such as the above-mentioned first consistency information and second consistency information
- the loss function of this model can be defined as: where ⁇ is the hyperparameter, ⁇ is all trainable parameters, N is the size of the training data in the dataset, is the actual reply message.
- the model is optimized based on the loss function. When the loss value of the loss function meets the preset threshold, it can be confirmed that the model training is completed, and the model parameters can be obtained based on the trained model.
- the structure of the system for determining target reply information may include:
- Context encoding module Given the overall above information and current candidate answer information, use the BERT model to encode the input information, obtain the word-level word vector representation of the given overall above information and current candidate answer information, and obtain the overall The semantic representation of the above information and the semantic representation of the candidate answer information.
- Fine-grained comparison module Given the overall above information and the word vector of candidate answer information, the total difference information between the current candidate answer information and other candidate answer information can be obtained by calculating the similarity matrix between the current candidate answer information and other candidates . Then, using the bi-phase matching mechanism (that is, the process of obtaining H h_i and H i_h ), the overall above-mentioned word vector and the above-mentioned total difference information are used in both positive and negative directions (that is, calculating the above representation related to the reply, and the above-mentioned Relevant replies indicate these two directions) to compare respectively to obtain the reply-aware above representation and the above-aware reply representation, and combine the two through the gate mechanism to obtain the overall above and the current candidate reply information.
- the bi-phase matching mechanism that is, the process of obtaining H h_i and H i_h
- the overall above-mentioned word vector and the above-mentioned total difference information are used in both positive and negative directions (that is, calculating the above
- Prediction module Given the semantic representation of the overall context information and the current candidate answer information, the first consistent information between the overall context information and the candidate answer information, the speaker's overall context information (ie, the target context information) The second consistency information between the current candidate reply information and the three are concatenated and input into a softmax network to obtain the final score of the candidate reply.
- the training objective of the model is to maximize the probability value of the correct response.
- the dialogue history information is "A: It's half past six, shall we start making dinner now? B: But I don't want to cook anymore, I'm too tired from cooking every day. A: Shall we go out to eat? On Third Avenue A new Chinese restaurant opened. Xiaoming went there yesterday and said it was delicious. B: Really? What type of food do they have? You know, I can’t eat spicy food. A: Don’t worry, the chef is from Cantonese .I know Cantonese food is one of your favorite dishes. B: Great, do you know how to get there? A: I don't know where, I just know it's on Third Avenue. Don't worry, I think we can Found it. B: But I don't want to walk, it's too hot outside. A: Then let Xiao Ming pick us up, we can treat him to dinner. B: Good idea".
- Candidate response information includes the following options: "1)Since Sichuan food is your favorite dish, why don't we go there after work? 2)Okay, I know where the restaurant is. 3)Okay, let's go! 4) Okay, we can go eat your favorite Sichuan food after school.”
- option 1) splice the overall above information with option 1), use the BERT model to obtain the vector representation of each word, and obtain the same-dimensional vector representation of the entire semantic information.
- option 1) and option 2) is [Sichuan food is your favorite dish, we will eat it after work]; the difference between option 1) and option 3) is [Sichuan food is your favorite dish]; option 1 ) and option 4) is [after work, after school].
- option 1) and option 2) is [Sichuan food is your favorite dish, we will eat it after work];
- option 1) and option 3) is [Sichuan food is your favorite dish]; option 1 ) and option 4) is [after work, after school].
- option 1) and option 3) is [Sichuan food is your favorite dish]; option 1 ) and option 4) is [after work, after school].
- the matching degree between the differential content and the historical context of speaker A is calculated, and it is found that speaker A has said that the other party’s favorite Cantonese food and other information, so the comparison between such information and differential content will be strengthened, and Sichuan Cuisine and Cantonese cuisine do not match, so option 1) has a relatively low speaker consistency score (the degree of consistency contained in the second consistency information is poor).
- the present disclosure provides an embodiment of a device for determining information, which is similar to the method embodiments shown in FIG. 2 and FIG. 3 Correspondingly, the device can be specifically applied to various electronic devices.
- the apparatus for determining information in this embodiment includes: an acquiring unit 601 , a first determining unit 602 , a second determining unit 603 , and a third determining unit 604 .
- the acquiring unit is configured to acquire the overall context information and a plurality of candidate answer information for replying the overall context information;
- the first determination unit is configured to, for each candidate answer information in the plurality of candidate answer information, Determine the total difference information between the candidate answer information and other candidate answer information except the candidate answer information among the plurality of candidate answer information;
- the second determination unit is configured to combine the total difference information with the overall above information The consistency information among them is determined as the first consistency information between the candidate answer information and the overall above information;
- the third determining unit is configured to, according to the first consistency information corresponding to each candidate answer information, Target answer information is determined from a plurality of candidate answer information.
- the first determination unit includes: a first acquisition module configured to acquire a word vector of each candidate answer information; a first determination module configured to, for each candidate answer information, based on the candidate answer The similarity between the word vector of the information and the word vector of each other candidate answer information in other candidate answer information, determine the candidate answer information and each Difference information between one other candidate answer information; the second determination module is configured to determine the difference information between the candidate answer information and all other answer information as the total of the candidate answer information and other candidate answer information diff information.
- the second determining unit includes: a second obtaining module configured to obtain a word vector of the overall above information; a third determining module configured to combine the total difference information with the word vector of the overall above information The consistency information between is determined as the first consistency information between the candidate answer information and the overall above information.
- the apparatus includes: a fourth determination unit configured to determine the consistency information between the total difference information and the target context information as the second one between the candidate reply information and the target context information Consistency information, wherein the target above information is the information in the overall above information that belongs to the same interlocutor as the candidate reply information; the third determination unit includes: a fourth determination module configured to Corresponding to the first consistency information and the second consistency information, target answer information is determined from multiple candidate answer information.
- the total difference information is determined based on the similarity between the word vector of the candidate answer information and the word vectors of other candidate answer information in the plurality of candidate answer information except the candidate answer information, and the fourth determination
- the unit includes: a third acquisition module configured to acquire the word vector of the target context information; a fifth determination module configured to determine the consistency information between the total difference information and the word vector of the target context information as The second consistency information between the candidate answer information and the target context information.
- the fourth determination module includes: a scoring module configured to, for each candidate answer information, adopt the first consistency information corresponding to the candidate answer information, the second consistency information corresponding to the candidate answer information and the third consistency information between the semantics of the candidate answer information and the semantics of the overall above information, determine the score of the candidate answer information; the selection module is configured to, according to the score of each candidate answer information, Target answer information is determined from a plurality of candidate answer information.
- Each unit in the above apparatus 600 corresponds to the steps in the method described with reference to FIG. 2 and FIG. 3 . Therefore, the operations, features and achievable technical effects described above for the method for determining information are also applicable to the device 600 and the units contained therein, and will not be repeated here.
- the present application also provides an electronic device and a readable storage medium.
- FIG. 7 it is a block diagram of an electronic device 700 according to a method for determining information according to an embodiment of the present application.
- Electronic device is intended to represent various forms of digital computers, such as laptops, desktops, workstations, personal digital assistants, servers, blade servers, mainframes, and other suitable computers.
- Electronic devices may also represent various forms of mobile devices, such as personal digital processing, cellular telephones, smart phones, wearable devices, and other similar computing devices.
- the components shown herein, their connections and relationships, and their functions, are by way of example only, and are not intended to limit implementations of the applications described and/or claimed herein.
- the electronic device includes: one or more processors 701, a memory 702, and interfaces for connecting various components, including high-speed interfaces and low-speed interfaces.
- the various components are interconnected using different buses and can be mounted on a common motherboard or otherwise as desired.
- the processor may process instructions executed within the electronic device, including instructions stored in or on the memory, to display graphical information of a GUI on an external input/output device such as a display device coupled to an interface.
- multiple processors and/or multiple buses may be used with multiple memories and multiple memories, if desired.
- multiple electronic devices may be connected, with each device providing some of the necessary operations (eg, as a server array, a set of blade servers, or a multi-processor system).
- a processor 701 is taken as an example.
- the memory 702 is a non-transitory computer-readable storage medium provided in this application.
- the memory stores instructions executable by at least one processor, so that the at least one processor executes the method for determining information provided in this application.
- the non-transitory computer-readable storage medium of the present application stores computer instructions, and the computer instructions are used to cause a computer to execute the method for determining information provided in the present application.
- the memory 702 can be used to store non-transitory software programs, non-transitory computer-executable programs and modules, such as the program instructions/modules corresponding to the method for determining information in the embodiment of the present application ( For example, the first acquiring unit 601 , the first determining unit 602 , the second determining unit 603 , and the third determining unit 604 shown in FIG. 6 ).
- the processor 701 executes various functional applications and data processing of the server by running the non-transitory software programs, instructions and modules stored in the memory 702, that is, implements the method for determining information in the above method embodiments.
- the memory 702 may include a program storage area and a data storage area, wherein the program storage area may An operating system, an application program required for at least one function are stored; the storage data area may store data created according to use of the electronic device for extracting video clips, and the like.
- the memory 702 may include a high-speed random access memory, and may also include a non-transitory memory, such as at least one magnetic disk storage device, a flash memory device, or other non-transitory solid-state storage devices.
- the storage 702 may optionally include storages that are remotely located relative to the processor 701, and these remote storages may be connected to electronic devices for extracting video clips through a network. Examples of the aforementioned networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.
- the electronic device used in the method for determining information may further include: an input device 703 , an output device 704 , and a bus 705 .
- the processor 701, the memory 702, the input device 703, and the output device 704 may be connected through a bus 705 or in other ways, and the connection through the bus 705 is taken as an example in FIG. 7 .
- the input device 703 can receive input numbers or character information, and generate key signal inputs related to user settings and function control of electronic equipment for extracting video clips, such as touch screens, keypads, mice, trackpads, touchpads, pointers, etc. input devices such as sticks, one or more mouse buttons, trackballs, joysticks, etc.
- the output device 704 may include a display device, an auxiliary lighting device (eg, LED), a tactile feedback device (eg, a vibration motor), and the like.
- the display device may include, but is not limited to, a liquid crystal display (LCD), a light emitting diode (LED) display, and a plasma display. In some implementations, the display device may be a touch screen.
- Various implementations of the systems and techniques described herein can be implemented in digital electronic circuitry, integrated circuit systems, application specific ASICs (application specific integrated circuits), computer hardware, firmware, software, and/or combinations thereof. These various embodiments may include being implemented in one or more computer programs executable and/or interpreted on a programmable system including at least one programmable processor, the programmable processor Can be special-purpose or general-purpose programmable processor, can receive data and instruction from storage system, at least one input device, and at least one output device, and transmit data and instruction to this storage system, this at least one input device, and this at least one output device an output device.
- machine-readable medium and “computer-readable medium” refer to media used to provide machine instructions and/or data to a computer Any computer program product, apparatus, and/or means (eg, magnetic disk, optical disk, memory, programmable logic device (PLD)) that programs a processor, including a machine-readable medium that receives machine instructions as machine-readable signals.
- machine-readable signal refers to any signal used to provide machine instructions and/or data to a programmable processor.
- the systems and techniques described herein can be implemented on a computer having a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user. ); and a keyboard and pointing device (eg, a mouse or a trackball) through which a user can provide input to the computer.
- a display device e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor
- a keyboard and pointing device eg, a mouse or a trackball
- Other kinds of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and can be in any form (including Acoustic input, speech input or, tactile input) to receive input from the user.
- the systems and techniques described herein can be implemented in a computing system that includes back-end components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes front-end components (e.g., as a a user computer having a graphical user interface or web browser through which a user can interact with embodiments of the systems and techniques described herein), or including such backend components, middleware components, Or any combination of front-end components in a computing system.
- the components of the system can be interconnected by any form or medium of digital data communication, eg, a communication network. Examples of communication networks include: Local Area Network (LAN), Wide Area Network (WAN) and the Internet.
- a computer system may include clients and servers.
- Clients and servers are generally remote from each other and typically interact through a communication network.
- the relationship of client and server arises by computer programs running on the respective computers and having a client-server relationship to each other.
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Databases & Information Systems (AREA)
- Mathematical Physics (AREA)
- Human Computer Interaction (AREA)
- Health & Medical Sciences (AREA)
- Audiology, Speech & Language Pathology (AREA)
- General Health & Medical Sciences (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
本申请公开了用于确定信息的方法和装置，涉及计算机技术领域。该方法包括：获取整体上文信息以及用于答复整体上文信息的多个候选答复信息；针对多个候选答复信息中的每一个候选答复信息，确定该候选答复信息、与多个候选答复信息中除该候选答复信息之外的其他候选答复信息之间的总差异信息；将总差异信息与整体上文信息之间的一致性信息，确定为该候选答复信息与整体上文信息之间的第一一致性信息；根据每一个候选答复信息对应的第一一致性信息，从多个候选答复信息中确定目标答复信息。
Description
本专利申请要求于2022年02月17日提交的、申请号为202210151466.6、发明名称为“用于确定信息的方法和装置”的中国专利申请的优先权，该申请的全文以引用的方式并入本申请中。
本公开涉及计算机技术领域，具体涉及用于确定信息的方法和装置。
随着人工智能技术的发展，越来越多的领域采用人工智能模型实现人机对话系统，例如，智能客服系统、智能聊天系统以及自助问答系统等。现有的人机对话系统通常基于历史对话中的整体上文信息的语义确定答复/回复信息。
然而，基于整体上文信息的语义确定答复/回复信息的方法，存在确定信息不准确的问题。
发明内容
本公开提供了一种用于确定信息的方法、装置、电子设备，计算机可读存储介质以及计算机程序产品。
本公开的一些实施例提供了一种用于确定信息的方法，包括：获取整体上文信息以及用于答复整体上文信息的多个候选答复信息；针对多个候选答复信息中的每一个候选答复信息，确定该候选答复信息、与多个候选答复信息中除该候选答复信息之外的其他候选答复信息之间的总差异信息；将总差异信息与整体上文信息之间的一致性信息，确定为该候选答复信息与整体上文信息之间的第一一致性信息；根据每一个候选答复信息对应的第一一致性信息，从多个候选答复信息中确定目标答复信息。
本公开的一些实施例提供了一种用于确定信息的装置，包括：获取单
元，被配置为获取整体上文信息以及用于答复整体上文信息的多个候选答复信息；第一确定单元，被配置为针对多个候选答复信息中的每一个候选答复信息，确定该候选答复信息、与多个候选答复信息中除该候选答复信息之外的其他候选答复信息之间的总差异信息；第二确定单元，被配置为将总差异信息与整体上文信息之间的一致性信息，确定为该候选答复信息与整体上文信息之间的第一一致性信息；第三确定单元，被配置为根据每一个候选答复信息对应的第一一致性信息，从多个候选答复信息中确定目标答复信息。
本公开的一些实施例提供了一种电子设备，包括：一个或多个处理器：存储装置，用于存储一个或多个程序，当一个或多个程序被一个或多个处理器执行，使得一个或多个处理器实现如第一方面提供的用于确定信息的方法。
本公开的一些实施例提供了一种计算机可读存储介质，其上存储有计算机程序，其中，程序被处理器执行时实现如上述描述的用于确定信息的方法的任一实施例。
本公开的一些实施例提供了一种计算机程序产品，包括计算机程序，所述计算机程序在被处理器执行时实现如上述描述的用于确定信息的方法的任一实施例。
应当理解，本部分所描述的内容并非旨在标识本公开的实施例的关键或重要特征，也不用于限制本公开的范围。本公开的其它特征将通过以下的说明书而变得容易理解。
附图用于更好地理解本方案，不构成对本申请的限定。其中：
图1是本申请的实施例可以应用于其中的示例性系统架构图；
图2是根据本申请的用于确定信息的方法的一个实施例的流程图；
图3是根据本申请的用于确定信息的方法的另一个实施例的流程图；
图4是根据本申请的用于确定信息的方法的应用场景的流程图；
图5是根据本申请的用于确定信息的方法中的应用场景中所采用的模型的训练装置示意图；
图6是根据本申请的用于确定信息的装置的一个实施例的结构示意图；
图7是用来实现本申请实施例的用于确定信息的方法的电子设备的框图。
以下结合附图对本申请的示范性实施例做出说明，其中包括本申请实施例的各种细节以助于理解，应当将它们认为仅仅是示范性的。因此，本领域普通技术人员应当认识到，可以对这里描述的实施例做出各种改变和修改，而不会背离本申请的范围和精神。同样，为了清楚和简明，以下的描述中省略了对公知功能和结构的描述。
图1示出了可以应用本申请的用于确定信息的方法或用于确定信息的装置的实施例的示例性系统架构100。
如图1所示，系统架构100可以包括终端设备101、102、103，网络104和服务器105。网络104用以在终端设备101、102、103和服务器105之间提供通信链路的介质。网络104可以包括各种连接类型，例如有线、无线通信链路或者光纤电缆等等。
用户可以使用终端设备101、102、103通过网络104与服务器105交互，以接收或发送消息等。终端设备101、102、103可以是用户终端设备，其上可以安装有各种客户端应用，例如图像类应用、视频类应用、购物类应用、聊天类应用、搜索类应用、金融类应用等。
终端设备101、102、103可以是具有显示屏并且支持接收服务器消息的各种电子设备，包括但不限于智能手机、平板电脑、电子书阅读器、电子播放器、膝上型便携计算机和台式计算机等等。
终端设备101、102、103可以是硬件，也可以是软件。当终端设备101、102、103为硬件时，可以是各种电子设备，当终端设备101、102、103为软件时，可以安装在上述所列举的电子设备中。其可以实现成多个软件或软件模块(例如用来提供分布式服务的多个软件模块)，也可以实现成单个软件或软件模块。在此不做具体限定。
服务器105可以获取整体上文信息以及用于答复整体上文信息的多个
候选答复信息后，针对多个候选答复信息中的每一个候选答复信息，确定该候选答复信息与多个候选答复信息之间的总差异信息，将该总差异信息与整体上文信息之间的一致性信息确定为该候选答复信息与整体上文信息之间的第一一致性信息，并根据每一个候选答复信息对应的第一一致性信息，从多个候选答复信息中确定目标答复信息。
需要说明的是，本公开的实施例所提供的用于确定信息的方法可以由服务器105执行，相应地，用于确定信息的装置可以设置于服务器105中。
应该理解，图1中的终端设备、网络和服务器的数目仅仅是示意性的。根据实现需要，可以具有任意数目的终端设备、网络和服务器。
继续参考图2，示出了根据本公开的用于确定信息的方法的一个实施例的流程200，包括以下步骤：
步骤201，获取整体上文信息以及用于答复整体上文信息的多个候选答复信息。
在本实施例中，用于确定信息的方法的执行主体(例如图1所示的服务器105)可以获取整体上文信息以及用于答复该整体上文信息的多个候选答复信息。其中，整体上文信息指是对话中的历史信息，如智能客服系统中的历史问答消息，智能聊天系统中的历史对话记录等。答复信息是指用于答复该整体上文信息中最后一条信息、或者全部信息的答复信息。
步骤202，针对多个候选答复信息中的每一个候选答复信息，确定该候选答复信息、与多个候选答复信息中除该候选答复信息之外的其他候选答复信息之间的总差异信息。
在本实施例中，针对多个候选答复信息中的每一个候选答复信息，可以确定出该条候选答复信息、与多个候选答复信息中除该候选答复信息外的其他候选答复信息之间的总差异信息。例如，若候选答复信息A为“今天天气晴朗”，候选答复信息B为“今天天气晴朗、湿度适中”，候选答复信息C为“今天天气晴朗、适合出行”，候选答复信息D为“今天天气晴朗，适合出行，适合去爬山”，则候选答复信息D、与多个候选答复信息中除候选答复信息外的其他候选答复信息之间的总差异信息为“适合出行、适合去爬山”。
步骤203，将总差异信息与整体上文信息之间的一致性信息，确定为
该候选答复信息与整体上文信息之间的第一一致性信息。
在本实施例中，可以基于该候选答复信息、与除该候选答复信息之外的其他候选答复信息之间的总差异信息，确定该候选答复信息与整体上文信息之间的一致性信息，为便于区分，该一致性信息可以称之为第一一致性信息，该一致性信息用于表征总差异信息所表达的内容、与整体上文信息所表达的内容之间的逻辑一致性程度/相似性程度，或者总差异信息的语法/语义、与整体上文信息的语法/语义之间的逻辑一致性程度/相似性程度等。
步骤204，根据每一个候选答复信息对应的第一一致性信息，从多个候选答复信息中确定目标答复信息。
在本实施例中，可以根据每一个候选答复信息对应的第一一致性信息，从多个候选答复信息中确定出目标答复信息，其中，与任意答复信息对应的第一一致性信息是指：该任意候选答复信息与其他候选答复信息之间的总差异信息、与整体上文信息之间的一致性信息。
本实施例提供的用于确定信息的方法，获取整体上文信息以及用于答复整体上文信息的多个候选答复信息；针对多个候选答复信息中的每一个候选答复信息，确定该候选答复信息、与多个候选答复信息中除该候选答复信息之外的其他候选答复信息之间的总差异信息；将总差异信息与整体上文信息之间的一致性信息，确定为该候选答复信息与整体上文信息之间的第一一致性信息；根据每一个候选答复信息对应的第一一致性信息，从多个候选答复信息中确定目标答复信息，可以增强所确定的目标答复信息与历史对话记录中的整体上文信息在逻辑以及内容上的一致性，从而提高推送信息的准确性。
可选地，针对多个候选答复信息中的每一个候选答复信息，确定该候选答复信息、与多个候选答复信息中除该候选答复信息之外的其他候选答复信息之间的总差异信息，包括：获取每一个候选答复信息的词向量；针对每一个候选答复信息，基于该候选答复信息的词向量、与其他候选答复信息中的每一个其他候选答复信息的词向量之间的相似度，确定该候选答复信息与每一个其他候选答复信息之间的差异信息；将该候选答复信息与全部其他答复信息之间的差异信息，确定为该候选答复信息与其他候选答
复信息之间的总差异信息。
在本实施例中，可以基于预先训练好的语义解析模型、或预先训练好的语言表征模型等获取每一个候选答复信息的词向量，以针对每一个候选答复信息，基于该候选答复信息的词向量、与多个候选答复中的每一个其他候选答复信息的词向量之间的相似度，确定该候选答复信息与每一个候选答复信息之间的差异信息，该差异信息可以是二者信息所表达的语义之间的差异内容、或者二者信息所表现的逻辑之间的差异信息，之后，将该候选答复信息与每一个其他候选答复信息之间的差异信息进行汇总/集合，得到该候选答复信息与全部的其他答复信息之间的总差异信息，也即，该候选答复信息、与多个候选答复信息中除该候选答复信息之外的其他候选答复信息之间的总差异信息。
本实施例基于候选答复信息的词向量之间的相似度，确定候选答复信息之间的差异信息，可以提高确定差异信息的准确性以及效率。
可选地，将总差异信息与整体上文信息之间的一致性信息，确定为该候选答复信息与整体上文信息之间的第一一致性信息，包括：获取整体上文信息的词向量；将总差异信息与整体上文信息的词向量之间的一致性信息，确定为该候选答复信息与整体上文信息之间的第一一致性信息。
在本实施例中，可以基于预先训练好的语义解析模型、或预先训练好的语言表征模型等获取整体上文信息的词向量，之后基于整体上文信息的词向量、以及该候选答复信息与其他候选答复信息之间的总差异信息，确定该候选答复信息与整体上文信息之间的第一一致性信息。可以理解，由于总差异信息是基于候选答复信息的词向量之间的差异所确定，所以总差异信息的数据格式为词向量形式，基于整体上文信息的词向量与总差异信息可以直接进行一致性对比，也即，相似性/相同性对比，以获得该候选信息与整体上文信息之间的一致性信息。
本实施例基于整体上文信息的词向量与总差异信息确定该候选答复信息与整体上文信息之间的第一一致性信息，可以提高确定第一一致性信息的准确性以及效率。
继续参考图3，示出了根据本公开的用于确定信息的方法的另一个实施例的流程300，包括以下步骤：
步骤301，获取整体上文信息以及用于答复整体上文信息的多个候选答复信息。
步骤302，针对多个候选答复信息中的每一个候选答复信息，确定该候选答复信息、与多个候选答复信息中除该候选答复信息之外的其他候选答复信息之间的总差异信息。
步骤303，将总差异信息与整体上文信息之间的一致性信息，确定为该候选答复信息与整体上文信息之间的第一一致性信息。
本实施例中对步骤301、步骤302、步骤303的描述与步骤201、步骤202、步骤203的描述一致，此处不再赘述。
步骤304，将总差异信息与目标上文信息之间的一致性信息，确定为该候选答复信息与目标上文信息之间的第二一致性信息，其中，目标上文信息为整体上文信息中的、与候选答复信息属于同一对话方的信息。
在本实施例中，可以将总差异信息与目标上文信息之间的一致性信息，确定为该候选答复信息与目标上文信息之间的一致性信息，为便于区分，该一致性信息可以称之为第二一致性信息，该一致性信息用于表征总差异信息所表达的内容、与目标上文信息所表达的内容之间的逻辑一致性，或者总差异信息的语法/语义、与目标上文信息的语法/语义之间的逻辑一致性等，目标上文信息是指：整体上文信息中、候选答复信息的对话方所生成的对话记录，例如，整体上文信息为A、B二者之间的对话(其中，A可以是真人用户，用于提出问题，B可以是机器人客服，用于回复用户的问题)，若当前对话进程处于B需要答复的阶段，即，当前需要由B生成消息，则候选答复信息的对话方为B，因此，目标上文信息为整体上文信息中所有来自B的消息。
步骤305，根据每一个候选答复信息对应的第一一致性信息与第二一致性信息，从多个候选答复信息中确定目标答复信息。
在本实施例中，可以根据每一个候选答复信息对应的第一一致性信息与第二一致性信息，从多个候选答复信息中确定出目标答复信息，其中，与任意答复信息对应的第一一致性信息是指：该任意候选答复信息与其他候选答复信息之间的总差异信息、与整体上文信息之间的一致性信息；与任意答复信息对应的第二一致性信息是指：该任意候选答复信息与其他候
选答复信息之间的总差异信息、与目标上文信息之间的一致性信息。
本实施例提供的用于确定信息的方法，相比于图2实施例描述的方法，增加了获取该候选答复信息与目标上文信息之间的第二一致性信息的步骤，以及在确定目标答复信息时还基于第二一致性信息，可以增强所确定的候选答复信息与历史对话记录中的目标上文信息在逻辑以及内容上的一致性，即增强候选答复信息对话方自身提供的信息的前后一致性，从而提高确定回复信息的准确性。
可选地，总差异信息基于该候选答复信息的词向量、与多个候选答复信息中除该候选答复信息之外的其他候选答复信息的词向量之间的相似度确定，将总差异信息与目标上文信息之间的一致性信息，确定为该候选答复信息与目标上文信息之间的第二一致性信息，包括：获取目标上文信息的词向量；将总差异信息与目标上文信息的词向量之间的一致性信息，确定为该候选答复信息与目标上文信息之间的第二一致性信息。
在本实施例中，可以基于预先训练好的语义解析模型、或预先训练好的语言表征模型等获取目标上文信息的词向量，之后基于目标上文信息的词向量、以及该候选答复信息与其他候选答复信息之间的总差异信息，确定该候选答复信息与目标上文信息之间的第二一致性信息。可以理解，由于总差异信息是基于该候选答复信息的词向量、与其他候选答复信息的词向量之间的差异所确定，所以总差异信息的数据格式为词向量形式，基于目标上文信息的词向量与总差异信息可以直接进行一致性对比，也即，相似性/相同性对比，以获得该候选信息与目标上文信息之间的一致性信息。
本实施例基于目标上文信息的词向量与总差异信息确定该候选答复信息与整体上文信息之间的第二一致性信息，可以提高确定第二一致性信息的准确性以及效率。
可选地，根据每一个候选答复信息对应的第一一致性信息与第二一致性信息，从多个候选答复信息中确定目标答复信息，包括：针对每一个候选答复信息，采用该候选答复信息对应的第一一致性信息、该候选答复信息对应的第二一致性信息、以及该候选答复信息的语义与整体上文信息的语义之间的第三一致性信息，确定该候选答复信息的得分；根据每一个候选答复信息的得分，从多个候选答复信息中确定目标答复信息。
在本实施例中，针对每一个候选答复信息，可以采用该候选答复信息对应的第一一致性信息、该候选答复信息对应的第二一致性信息、以及该候选答复信息的语义与整体上文信息的语义之间的第三一致性、，确定该候选答复信息的得分，具体地，使该候选答复信息的得分与上文语义相关性(上文语义相关性是指：该候选答复信息的语义与整体上文信息之间的语义之间的相关性程度/一致性程度)呈正比、与第一一致性信息呈正比(第一一致性程度越高的候选答复信息的得分越高)、与第二一致性信息呈正比(第二一致性程度越高的候选答复信息的得分越高)，在确定出每一个候选答复信息的得分后，将得分最高的候选答复信息确定为目标答复信息。
本实施例中，将上文语义相关性越大、第一一致性程度越高以及第二一致性程度越高的候选答复信息确定为目标答复信息，可以提高确定目标答复信息的准确性。
在一些应用场景中，如图4所示，用于确定信息的方法包括：
第一步，将历史对话/整体上文信息、以及用于答复该整体上文信息的多个候选答复信息输入BERT(Bidirectional Encoder Representations from Transformer，基于转换器的双向编码表征)模型中，以获得BERT模型输出的整体上文信息的词向量，以及多个候选答复信息中的每一个候选答复信息的词向量。
具体包括：获取(U；ri)，其中U＝{u1，...，uN}表示整体上文信息，代表第i句上文语句，代表第i句整体上文信息中的第N个词语，代表第i个候选答复信息，代表第i句候选答复信息中的第N个词语，将整体上文信息U以及每个候选答复信息ri连接后，输入预先训练的BERT模型中，并获得BERT模型输出的整体上文信息的词向量以及候选答复信息的词向量：
其中，BERT(·)返回BERT模型的最后一层输出。<；>代表分号前后
两个序列的串联/拼接，HU代表整体上文信息U的词向量，代表候选答复信息ri的词向量。另外，基于BERT模型还可以获得整体上文信息的语义信息的总向量
其中，BERT(·)返回BERT模型的最后一层输出。<；>代表分号前后
两个序列的串联/拼接，HU代表整体上文信息U的词向量，代表候选答复信息ri的词向量。另外，基于BERT模型还可以获得整体上文信息的语义信息的总向量
第二步，在获得整体上文信息的词向量以及候选答复信息的词向量后，基于二者进行细粒度回复对比。
具体包括：首先，针对每一个候选答复信息，计算整体上文信息的词向量HU与该候选答复信息的词向量之间的词级注意力，并得到二者之间的相似矩阵：
其中，
其中，⊙代表对两个矩阵之间进行元素乘法。是ri中第m个词语的
向量表示，是模型参数。表示ri的第m个单词和rj的第n个单词之间的相似性。
其中，
其中，⊙代表对两个矩阵之间进行元素乘法。是ri中第m个词语的
向量表示，是模型参数。表示ri的第m个单词和rj的第n个单词之间的相似性。
其次，基于上述相似矩阵得到ri与rj之间的差异信息，ri与rj的对比信息定义为：
其中，表示代表ri与rj之间相似的部分，表示ri与rj之间不同的部分，
即，ri与rj之间的差异信息。
其中，表示代表ri与rj之间相似的部分，表示ri与rj之间不同的部分，
即，ri与rj之间的差异信息。
最后，在获取到当前ri与多个候选答复信息中的每一个rj之间的差异信息后，综合上述差异信息，得到ri与其他所有候选答复信息的差异化表示即总差异信息：
其中，是计算式的中间变量；[·]代表拼接操作，即将[]中的序列
进行拼接；σ(·)代表狄拉克δ函数；W2、W3、b2代表模型参数。
其中，是计算式的中间变量；[·]代表拼接操作，即将[]中的序列
进行拼接；σ(·)代表狄拉克δ函数；W2、W3、b2代表模型参数。
第三步，进行整体上文逻辑的一致性推理：基于ri与其他所有候选答复信息的差异化表示确定该候选答复信息ri与整体上文信息(具体为词向量HU)之间的一致性信息(即，上文中所称的第一一致性信息)。
具体包括：
Hi_h＝Relu(Ai_hHUW7)
Eh_i＝MaxPolling(Hh_i)
Ei_h＝MaxPolling(Hi_h)
ghi＝σ(Eh_iW8+Ei_hW9+b4)
其中W4、W5、W6、W7、W8、W9、b4是模型参数；Ah_i和Ai_h是全部整体
上文信息U与候选答复信息ri之间的单词级别的注意力矩阵，它们关注不同的视角，即，求取两种数据所采用的注意力模型关注的信息不同，求取Ah_i的所采用的的注意力模型关注的是与回复信息相关的整体上文信息，求取Ai_h所采用的注意力模型关注的是与整体上文信息相关的回复信息。MaxPooling代表最大池操作，SoftMax(·)代表一种归一化指数函数，Relu(·)代表一种线性整流函数；Hh_i是回复感知的上文表示，Hi_h是上文感知的回复表示；ghi表示将Hh_i和Hi_h基于门机制融合后得到的门机制值。
其中W4、W5、W6、W7、W8、W9、b4是模型参数；Ah_i和Ai_h是全部整体
上文信息U与候选答复信息ri之间的单词级别的注意力矩阵，它们关注不同的视角，即，求取两种数据所采用的注意力模型关注的信息不同，求取Ah_i的所采用的的注意力模型关注的是与回复信息相关的整体上文信息，求取Ai_h所采用的注意力模型关注的是与整体上文信息相关的回复信息。MaxPooling代表最大池操作，SoftMax(·)代表一种归一化指数函数，Relu(·)代表一种线性整流函数；Hh_i是回复感知的上文表示，Hi_h是上文感知的回复表示；ghi表示将Hh_i和Hi_h基于门机制融合后得到的门机制值。
第四步，对话方整体信息一致性增强：基于ri与其他所有候选答复信息的差异化表示确定该候选答复信息ri与目标上文信息(具体为词向量HS)之间的一致性信息(即，上文中所称的第二一致性信息)，其中，目标上文信息是指该候选答复信息所属对话方所给出的整体上文信息(例如，候选答复信息为用户A需要进行的对话信息，则目标上文信息是指整体上文信息中，所有用户A发出的信息)
Hi_s＝Relu(Ai_hHuW13)
其中，As_i和Ai_s是目标上文信息(即，对话方自身整体上文信息)与候选答复信息ri之间的单词级别的注意力矩阵，它们关注不同的视角，即，求取两种数据所采用的注意力模型关注的信息不同，求取As_i所采用的注意力模型关注的是目标上文信息，求取Ai_s所采用的注意力模型关注的是与目标上文信息相关的回复信息(答复信息)；标记T代表矩阵的转置；Hs_i代表回复感知的说话人表示，是整体上文信息的向量表示，这个向量表示与回复信息有关；Hi_s代表说话人感知的回复表示，是回复的向量表示，这个向量与目标上文信息相关；W10、W11、W12、W13是模型参数。
之后，可以确定候选答复信息ri与目标上文信息之间的一致性信息
Es_i＝MaxPolling(Hs_i)
Ei_s＝MaxPolling(Hi_s)
Es_i＝MaxPolling(Hs_i)
Ei_s＝MaxPolling(Hi_s)
gsi＝a(Es_iW14+Ei_sW15+b5)
其中，W14、W15、b5、b6是模型参数；gsi表示将Hs_i和Hi_s基于门机
制融合后得到的门机制值。
其中，W14、W15、b5、b6是模型参数；gsi表示将Hs_i和Hi_s基于门机
制融合后得到的门机制值。
第五步，首先，将整体上文信息的语义的总向量该候选答复信息ri与整体上文信息之间的一致性信息该候选答复信息ri与目标上文信息之间的一致性信息该三者数据进行拼接，以获得该候选答复信息ri的推理信息Hi：
其次，基于推理信息，预测候选答复信息ri的得分P(ri|U，R)为
其中，W16和b6是模型参数，集合M是多个候选答复信息中除该候选答复信息ri之外的候选答复信息的标识；R代表候选答复信息的集合。
在该应用场景中，上述模型参数可以基于模型训练得到，用于训练模型的系统可以如图5所示，训练模型的步骤包括：获取样本对话数据，其中包括对话历史以及候选回复，对样本对话数据进行过滤，以过滤掉不完整数据以及冗余数据，从而获得训练数据；将训练数据输入知识转移预测器以获得每条训练数据的词向量，将经过知识转移预测器处理后的训练数据输入知识感知生成器以获得知识感知信息(如上述第一一致性信息、第二一致性信息)，基于知识感知信息确定出目标答复信息。该模型的损失函数可以定义为：
其中λ是超参数，θ是所有可训练参数，N是数据集中训练数据的大小，是实际答复信息。基于损失函数对模型进行优化，当损失函数的损失值满足预设阈值后，可以确认模型训练完成，基于训练完成的模型可以获得模型参数。
其中λ是超参数，θ是所有可训练参数，N是数据集中训练数据的大小，是实际答复信息。基于损失函数对模型进行优化，当损失函数的损失值满足预设阈值后，可以确认模型训练完成，基于训练完成的模型可以获得模型参数。
在该应用场景中，用于确定目标答复信息的系统的结构可以包括：
上下文编码模块：给定整体上文信息和当前候选答复信息，使用BERT模型对输入的信息做编码处理，获得给定整体上文信息与当前候选答复信息的单词级别的词向量表示，以及获得整体上文信息的语义表示以及候选答复信息的语义表示。
细粒度对比模块：给定整体上文信息和候选答复信息的词向量，通过计算当前候选答复信息与其他候选的相似性矩阵，可以获得当前候选答复信息与其他候选答复信息之间的总差异信息。然后利用双相匹配机制(即，获得Hh_i与Hi_h的过程)，将整体上文词向量与上述总差异信息从正反两个方向(即，计算回复相关的上文表示、和与上文相关的回复表示这两个方向)分别进行比较，获得回复感知的上文表示和上文感知的回复表示，并通过门机制将二者结合，获得整体上文与当前候选答复信息之间的第一
一致性信息。
利用另一个双相匹配机制(即，获得Hs_i与Hi_s的过程)，将说话人的上文词向量与上述总差异信息进行对比，获得回复感知的说话人上文表示和说话人感知的回复表示，并通过门机制将二者结合，获得说话人上文与当前候选答复信息之间的第二一致性信息。
预测模块：给定整体上文信息与当前候选答复信息的语义表示、整体上文信息与候选答复信息之间的第一一致性信息、说话人整体上文信息(即，目标上文信息)与当前候选答复信息之间的第二一致性信息，通过将三者进行拼接输入到一个softmax网络中，以此获得候选回复的最终得分。模型的训练目标是最大化正确回复的概率值。
该应用场景中，用于确定信息的方法的一个具体示例为：
对话历史信息是“A：已经六点半了，我们现在开始准备晚饭吧？B：但是我不想做饭了，我每天都做饭太累了。A：我们出去吃吧？在第三大道上新开了一家中式餐厅。小明昨天去了，说非常好吃。B：真的吗？他们都有什么类型的食物？你知道的，我不能吃辣。A：别担心，这个厨师是广东人。我知道广东菜是你最喜欢的菜品之一。B：太棒了，你知道怎么去吗？A:我不知道在哪，我只知道在第三大道上。别担心，我觉得我们可以找到的。B:但是我不想走路，外面太热了。A:那要不让小明接我们过去，我们可以请他吃饭。B：好主意”。
候选答复信息包括以下选项：“1)既然四川菜是你最喜欢的菜品，为什么我们不下了班就过去呢？2)好的，我知道饭店在哪。3)好的，走吧！4)好的，放学后我们可以去吃你最喜欢的四川菜。”
以选项1)为例，将整体上文信息与选项1)进行拼接，使用BERT模型，获得每个单词的向量表示，并获得整个语义信息的同维度向量表示。
对比选项1)与选项2)的差别是[四川菜是你最喜欢的菜品，我们下班去吃]；选项1)与选项3)的差别是[四川菜是你最喜欢的菜品]；选项1)与选项4)的差别是[下班过去，放学后]。然后将差异性的内容进行整合，获得选项1)与全部其他选项之间的总差异信息[四川菜是你最喜欢的菜品，我们下班去吃]。将该总差异信息与整体上文信息进行匹配度计算，发现四川菜和广东菜并不匹配，文中也没有下班的词语，因此选项1)在整体上
文一致性方面得分会相对较低(第一一致性信息所包含的一致性程度差)。最后将差异性的内容与说话人A的历史上文进行匹配度计算，发现说话人A曾说过对方最喜欢的广东菜等信息，因此会加强这类信息与差异性内容的比较，发现四川菜和广东菜并不匹配，因此选项1)的说话人上文一致性得分也相对较低(第二一致性信息所包含的一致性程度差)。
同上，对比选项2)与选项1)3)4)的差异化内容，并计算匹配得分，由于“我知道饭店在哪”与原文A曾说过“不知道在哪但是相信可以找到”相矛盾，因此匹配度得分较低(选项的语义与整体上文的语义之间的一致性差)。对比选项3)与选项1)2)4)，由于没有矛盾信息，因此相对分数较高。对比选项4)与选项1)2)3)，并计算匹配得分，由于“最喜欢四川菜”与原文A曾说过“你最喜欢的广东菜”相矛盾，因此匹配得分较低。最终将四个选项得分进行排序，确定选项3)是目标答复信息。
进一步参考图6，作为对上述各图所示方法的实现，本公开提供了一种用于确定信息的装置的一个实施例，该装置实施例与图2和图3所示的方法实施例相对应，该装置具体可以应用于各种电子设备中。
如图6所示，本实施例的用于确定信息的装置，包括：获取单元601、第一确定单元602、第二确定单元603、第三确定单元604。其中，获取单元，被配置为获取整体上文信息以及用于答复整体上文信息的多个候选答复信息；第一确定单元，被配置为针对多个候选答复信息中的每一个候选答复信息，确定该候选答复信息、与多个候选答复信息中除该候选答复信息之外的其他候选答复信息之间的总差异信息；第二确定单元，被配置为将总差异信息与整体上文信息之间的一致性信息，确定为该候选答复信息与整体上文信息之间的第一一致性信息；第三确定单元，被配置为根据每一个候选答复信息对应的第一一致性信息，从多个候选答复信息中确定目标答复信息。
在一些实施例中，第一确定单元，包括：第一获取模块，被配置为获取每一个候选答复信息的词向量；第一确定模块，被配置为针对每一个候选答复信息，基于该候选答复信息的词向量、与其他候选答复信息中的每一个其他候选答复信息的词向量之间的相似度，确定该候选答复信息与每
一个其他候选答复信息之间的差异信息；第二确定模块，被配置为将该候选答复信息与全部其他答复信息之间的差异信息，确定为该候选答复信息与其他候选答复信息之间的总差异信息。
在一些实施例中，第二确定单元，包括：第二获取模块，被配置为获取整体上文信息的词向量；第三确定模块，被配置为将总差异信息与整体上文信息的词向量之间的一致性信息，确定为该候选答复信息与整体上文信息之间的第一一致性信息。
在一些实施例中，装置包括：第四确定单元，被配置为将总差异信息与目标上文信息之间的一致性信息，确定为该候选答复信息与目标上文信息之间的第二一致性信息，其中，目标上文信息为整体上文信息中的、与候选答复信息属于同一对话方的信息；第三确定单元，包括：第四确定模块，被配置为根据每一个候选答复信息对应的第一一致性信息与第二一致性信息，从多个候选答复信息中确定目标答复信息。
在一些实施例中，总差异信息基于该候选答复信息的词向量、与多个候选答复信息中除该候选答复信息之外的其他候选答复信息的词向量之间的相似度确定，第四确定单元，包括：第三获取模块，被配置为获取目标上文信息的词向量；第五确定模块，被配置为将总差异信息与目标上文信息的词向量之间的一致性信息，确定为该候选答复信息与目标上文信息之间的第二一致性信息。
在一些实施例中，第四确定模块，包括：评分模块，被配置为针对每一个候选答复信息，采用该候选答复信息对应的第一一致性信息、该候选答复信息对应的第二一致性信息、以及该候选答复信息的语义与整体上文信息的语义之间的第三一致性信息，确定该候选答复信息的得分；选择模块，被配置为根据每一个候选答复信息的得分，从多个候选答复信息中确定目标答复信息。
上述装置600中的各单元与参考图2和图3描述的方法中的步骤相对应。由此上文针对用于确定信息的方法描述的操作、特征及所能达到的技术效果同样适用于装置600及其中包含的单元，在此不再赘述。
根据本申请的实施例，本申请还提供了一种电子设备和一种可读存储介质。
如图7所示，是根据本申请实施例的用于确定信息的方法的电子设备700的框图。电子设备旨在表示各种形式的数字计算机，诸如，膝上型计算机、台式计算机、工作台、个人数字助理、服务器、刀片式服务器、大型计算机、和其它适合的计算机。电子设备还可以表示各种形式的移动装置，诸如，个人数字处理、蜂窝电话、智能电话、可穿戴设备和其它类似的计算装置。本文所示的部件、它们的连接和关系、以及它们的功能仅仅作为示例，并且不意在限制本文中描述的和/或者要求的本申请的实现。
如图7所示，该电子设备包括：一个或多个处理器701、存储器702，以及用于连接各部件的接口，包括高速接口和低速接口。各个部件利用不同的总线互相连接，并且可以被安装在公共主板上或者根据需要以其它方式安装。处理器可以对在电子设备内执行的指令进行处理，包括存储在存储器中或者存储器上以在外部输入/输出装置(诸如，耦合至接口的显示设备)上显示GUI的图形信息的指令。在其它实施方式中，若需要，可以将多个处理器和/或多条总线与多个存储器和多个存储器一起使用。同样，可以连接多个电子设备，各个设备提供部分必要的操作(例如，作为服务器阵列、一组刀片式服务器、或者多处理器系统)。图7中以一个处理器701为例。
存储器702即为本申请所提供的非瞬时计算机可读存储介质。其中，该存储器存储有可由至少一个处理器执行的指令，以使该至少一个处理器执行本申请所提供的用于确定信息的方法。本申请的非瞬时计算机可读存储介质存储计算机指令，该计算机指令用于使计算机执行本申请所提供的用于确定信息的方法。
存储器702作为一种非瞬时计算机可读存储介质，可用于存储非瞬时软件程序、非瞬时计算机可执行程序以及模块，如本申请实施例中的用于确定信息的方法对应的程序指令/模块(例如，附图6所示的第获取单元601、第一确定单元602、第二确定单元603、第三确定单元604)。处理器701通过运行存储在存储器702中的非瞬时软件程序、指令以及模块，从而执行服务器的各种功能应用以及数据处理，即实现上述方法实施例中的用于确定信息的方法。
存储器702可以包括存储程序区和存储数据区，其中，存储程序区可
存储操作系统、至少一个功能所需要的应用程序；存储数据区可存储根据用于提取视频片段的电子设备的使用所创建的数据等。此外，存储器702可以包括高速随机存取存储器，还可以包括非瞬时存储器，例如至少一个磁盘存储器件、闪存器件、或其他非瞬时固态存储器件。在一些实施例中，存储器702可选包括相对于处理器701远程设置的存储器，这些远程存储器可以通过网络连接至用于提取视频片段的电子设备。上述网络的实例包括但不限于互联网、企业内部网、局域网、移动通信网及其组合。
用于确定信息的方法的电子设备还可以包括：输入装置703、输出装置704以及总线705。处理器701、存储器702、输入装置703和输出装置704可以通过总线705或者其他方式连接，图7中以通过总线705连接为例。
输入装置703可接收输入的数字或字符信息，以及产生与用于提取视频片段的电子设备的用户设置以及功能控制有关的键信号输入，例如触摸屏、小键盘、鼠标、轨迹板、触摸板、指示杆、一个或者多个鼠标按钮、轨迹球、操纵杆等输入装置。输出装置704可以包括显示设备、辅助照明装置(例如，LED)和触觉反馈装置(例如，振动电机)等。该显示设备可以包括但不限于，液晶显示器(LCD)、发光二极管(LED)显示器和等离子体显示器。在一些实施方式中，显示设备可以是触摸屏。
此处描述的系统和技术的各种实施方式可以在数字电子电路系统、集成电路系统、专用ASIC(专用集成电路)、计算机硬件、固件、软件、和/或它们的组合中实现。这些各种实施方式可以包括：实施在一个或者多个计算机程序中，该一个或者多个计算机程序可在包括至少一个可编程处理器的可编程系统上执行和/或解释，该可编程处理器可以是专用或者通用可编程处理器，可以从存储系统、至少一个输入装置、和至少一个输出装置接收数据和指令，并且将数据和指令传输至该存储系统、该至少一个输入装置、和该至少一个输出装置。
这些计算程序(也称作程序、软件、软件应用、或者代码)包括可编程处理器的机器指令，并且可以利用高级过程和/或面向对象的编程语言、和/或汇编/机器语言来实施这些计算程序。如本文使用的，术语“机器可读介质”和“计算机可读介质”指的是用于将机器指令和/或数据提供给可
编程处理器的任何计算机程序产品、设备、和/或装置(例如，磁盘、光盘、存储器、可编程逻辑装置(PLD))，包括，接收作为机器可读信号的机器指令的机器可读介质。术语“机器可读信号”指的是用于将机器指令和/或数据提供给可编程处理器的任何信号。
为了提供与用户的交互，可以在计算机上实施此处描述的系统和技术，该计算机具有：用于向用户显示信息的显示装置(例如，CRT(阴极射线管)或者LCD(液晶显示器)监视器)；以及键盘和指向装置(例如，鼠标或者轨迹球)，用户可以通过该键盘和该指向装置来将输入提供给计算机。其它种类的装置还可以用于提供与用户的交互；例如，提供给用户的反馈可以是任何形式的传感反馈(例如，视觉反馈、听觉反馈、或者触觉反馈)；并且可以用任何形式(包括声输入、语音输入或者、触觉输入)来接收来自用户的输入。
可以将此处描述的系统和技术实施在包括后台部件的计算系统(例如，作为数据服务器)、或者包括中间件部件的计算系统(例如，应用服务器)、或者包括前端部件的计算系统(例如，具有图形用户界面或者网络浏览器的用户计算机，用户可以通过该图形用户界面或者该网络浏览器来与此处描述的系统和技术的实施方式交互)、或者包括这种后台部件、中间件部件、或者前端部件的任何组合的计算系统中。可以通过任何形式或者介质的数字数据通信(例如，通信网络)来将系统的部件相互连接。通信网络的示例包括：局域网(LAN)、广域网(WAN)和互联网。
计算机系统可以包括客户端和服务器。客户端和服务器一般远离彼此并且通常通过通信网络进行交互。通过在相应的计算机上运行并且彼此具有客户端-服务器关系的计算机程序来产生客户端和服务器的关系。
应该理解，可以使用上面所示的各种形式的流程，重新排序、增加或删除步骤。例如，本申请中记载的各步骤可以并行地执行也可以顺序地执行也可以不同的次序执行，只要能够实现本申请公开的技术方案所期望的结果，本文在此不进行限制。
上述具体实施方式，并不构成对本申请保护范围的限制。本领域技术人员应该明白的是，根据设计要求和其他因素，可以进行各种修改、组合、子组合和替代。任何在本申请的精神和原则之内所作的修改、等同替换和
改进等，均应包含在本申请保护范围之内。
Claims (15)
- 一种用于确定信息的方法，包括：获取整体上文信息以及用于答复所述整体上文信息的多个候选答复信息；针对所述多个候选答复信息中的每一个候选答复信息，确定该候选答复信息、与所述多个候选答复信息中除该候选答复信息之外的其他候选答复信息之间的总差异信息；将所述总差异信息与所述整体上文信息之间的一致性信息，确定为该候选答复信息与所述整体上文信息之间的第一一致性信息；根据所述每一个候选答复信息对应的第一一致性信息，从所述多个候选答复信息中确定目标答复信息。
- 根据权利要求1所述的方法，其中，所述针对所述多个候选答复信息中的每一个候选答复信息，确定该候选答复信息、与所述多个候选答复信息中除该候选答复信息之外的其他候选答复信息之间的总差异信息，包括：获取所述每一个候选答复信息的词向量；针对所述每一个候选答复信息，基于该候选答复信息的词向量、与所述其他候选答复信息中的每一个其他候选答复信息的词向量之间的相似度，确定该候选答复信息与所述每一个其他候选答复信息之间的差异信息；将该候选答复信息与全部所述其他答复信息之间的差异信息，确定为该候选答复信息与所述其他候选答复信息之间的总差异信息。
- 根据权利要求2所述的方法，其中，所述将所述总差异信息与所述整体上文信息之间的一致性信息，确定为该候选答复信息与所述整体上文信息之间的第一一致性信息，包括：获取所述整体上文信息的词向量；将所述总差异信息与所述整体上文信息的词向量之间的一致性信息， 确定为该候选答复信息与所述整体上文信息之间的第一一致性信息。
- 根据权利要求1-3任一项所述的方法，其中，所述方法还包括：将所述总差异信息与目标上文信息之间的一致性信息，确定为该候选答复信息与所述目标上文信息之间的第二一致性信息，其中，所述目标上文信息为所述整体上文信息中的、与所述候选答复信息属于同一对话方的信息；所述根据所述每一个候选答复信息对应的第一一致性信息，从所述多个候选答复信息中确定目标答复信息，包括：根据所述每一个候选答复信息对应的第一一致性信息与第二一致性信息，从所述多个候选答复信息中确定所述目标答复信息。
- 根据权利要求4所述的方法，其中，所述总差异信息基于该候选答复信息的词向量、与所述多个候选答复信息中除该候选答复信息之外的其他候选答复信息的词向量之间的相似度确定；所述将所述总差异信息与目标上文信息之间的一致性信息，确定为该候选答复信息与所述目标上文信息之间的第二一致性信息，包括：获取所述目标上文信息的词向量；将所述总差异信息与所述目标上文信息的词向量之间的一致性信息，确定为该候选答复信息与所述目标上文信息之间的第二一致性信息。
- 根据权利要求4所述的方法，其中，所述根据所述每一个候选答复信息对应的第一一致性信息与第二一致性信息，从所述多个候选答复信息中确定所述目标答复信息，包括：针对所述每一个候选答复信息，采用该候选答复信息对应的第一一致性信息、该候选答复信息对应的第二一致性信息、以及该候选答复信息的语义与所述整体上文信息的语义之间的第三一致性信息，确定该候选答复信息的得分；根据每一个候选答复信息的得分，从所述多个候选答复信息中确定所述目标答复信息。
- 一种用于确定信息的装置，包括：获取单元，被配置为获取整体上文信息以及用于答复所述整体上文信息的多个候选答复信息；第一确定单元，被配置为针对所述多个候选答复信息中的每一个候选答复信息，确定该候选答复信息、与所述多个候选答复信息中除该候选答复信息之外的其他候选答复信息之间的总差异信息；第二确定单元，被配置为将所述总差异信息与所述整体上文信息之间的一致性信息，确定为该候选答复信息与所述整体上文信息之间的第一一致性信息；第三确定单元，被配置为根据所述每一个候选答复信息对应的第一一致性信息，从所述多个候选答复信息中确定目标答复信息。
- 根据权利要求7所述的装置，其中，所述第一确定单元，包括：第一获取模块，被配置为获取所述每一个候选答复信息的词向量；第一确定模块，被配置为针对所述每一个候选答复信息，基于该候选答复信息的词向量、与所述其他候选答复信息中的每一个其他候选答复信息的词向量之间的相似度，确定该候选答复信息与所述每一个其他候选答复信息之间的差异信息；第二确定模块，被配置为将该候选答复信息与全部所述其他答复信息之间的差异信息，确定为该候选答复信息与所述其他候选答复信息之间的总差异信息。
- 根据权利要求7所述的装置，其中，所述第二确定单元，包括：第二获取模块，被配置为获取所述整体上文信息的词向量；第三确定模块，被配置为将所述总差异信息与所述整体上文信息的词向量之间的一致性信息，确定为该候选答复信息与所述整体上文信息之间的第一一致性信息。
- 根据权利要求7-9任一项所述的装置，其中，所述装置包括：第四确定单元，被配置为将所述总差异信息与目标上文信息之间的一致性信息，确定为该候选答复信息与所述目标上文信息之间的第二一致性信息，其中，所述目标上文信息为所述整体上文信息中的、与所述候选答复信息属于同一对话方的信息；所述第三确定单元，包括：第四确定模块，被配置为根据所述每一个候选答复信息对应的第一一致性信息与第二一致性信息，从所述多个候选答复信息中确定所述目标答复信息。
- 根据权利要求10所述的装置，其中，所述总差异信息基于该候选答复信息的词向量、与所述多个候选答复信息中除该候选答复信息之外的其他候选答复信息的词向量之间的相似度确定，所述第四确定单元，包括：第三获取模块，被配置为获取所述目标上文信息的词向量；第五确定模块，被配置为将所述总差异信息与所述目标上文信息的词向量之间的一致性信息，确定为该候选答复信息与所述目标上文信息之间的第二一致性信息。
- 根据权利要求10所述的装置，其中，所述第四确定模块，包括：评分模块，被配置为针对所述每一个候选答复信息，采用该候选答复信息对应的第一一致性信息、该候选答复信息对应的第二一致性信息、以及该候选答复信息的语义与所述整体上文信息的语义之间的第三一致性信息，确定该候选答复信息的得分；选择模块，被配置为根据每一个候选答复信息的得分，从所述多个候选答复信息中确定所述目标答复信息。
- 一种电子设备，包括：至少一个处理器；以及与所述至少一个处理器通信连接的存储器；其中，所述存储器存储有可被所述至少一个处理器执行的指令，所述指令被 所述至少一个处理器执行，以使所述至少一个处理器能够执行权利要求1-6中任一项所述的方法。
- 一种存储有计算机指令的非瞬时计算机可读存储介质，其中，所述计算机指令用于使所述计算机执行权利要求1-6中任一项所述的方法。
- 一种计算机程序产品，包括计算机程序，所述计算机程序在被处理器执行时实现根据权利要求1-6中任一项所述的方法。
Applications Claiming Priority (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210151466.6A CN114547244A (zh) | 2022-02-17 | 2022-02-17 | 用于确定信息的方法和装置 |
| CN202210151466.6 | 2022-02-17 |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| WO2023155678A1 true WO2023155678A1 (zh) | 2023-08-24 |
Family
ID=81675380
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| PCT/CN2023/073922 WO2023155678A1 (zh) | 2022-02-17 | 2023-01-31 | 用于确定信息的方法和装置 |
Country Status (2)
| Country | Link |
|---|---|
| CN (1) | CN114547244A (zh) |
| WO (1) | WO2023155678A1 (zh) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114547244A (zh) * | 2022-02-17 | 2022-05-27 | 北京沃东天骏信息技术有限公司 | 用于确定信息的方法和装置 |
| CN115412745B (zh) * | 2022-08-12 | 2024-02-27 | 联想(北京)有限公司 | 一种信息处理方法及电子设备 |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20100114929A1 (en) * | 2008-11-06 | 2010-05-06 | Yahoo! Inc. | Diverse query recommendations using clustering-based methodology |
| CN108875074A (zh) * | 2018-07-09 | 2018-11-23 | 北京慧闻科技发展有限公司 | 基于交叉注意力神经网络的答案选择方法、装置和电子设备 |
| CN112948563A (zh) * | 2021-04-13 | 2021-06-11 | 天津禄智技术有限公司 | 文本搜索方法及其系统 |
| CN113392321A (zh) * | 2021-06-02 | 2021-09-14 | 北京三快在线科技有限公司 | 一种信息推荐方法、装置、电子设备及存储介质 |
| CN113703883A (zh) * | 2021-03-31 | 2021-11-26 | 腾讯科技（深圳）有限公司 | 一种交互方法和相关装置 |
| CN114547244A (zh) * | 2022-02-17 | 2022-05-27 | 北京沃东天骏信息技术有限公司 | 用于确定信息的方法和装置 |
-
2022
- 2022-02-17 CN CN202210151466.6A patent/CN114547244A/zh active Pending
-
2023
- 2023-01-31 WO PCT/CN2023/073922 patent/WO2023155678A1/zh active Application Filing
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20100114929A1 (en) * | 2008-11-06 | 2010-05-06 | Yahoo! Inc. | Diverse query recommendations using clustering-based methodology |
| CN108875074A (zh) * | 2018-07-09 | 2018-11-23 | 北京慧闻科技发展有限公司 | 基于交叉注意力神经网络的答案选择方法、装置和电子设备 |
| CN113703883A (zh) * | 2021-03-31 | 2021-11-26 | 腾讯科技（深圳）有限公司 | 一种交互方法和相关装置 |
| CN112948563A (zh) * | 2021-04-13 | 2021-06-11 | 天津禄智技术有限公司 | 文本搜索方法及其系统 |
| CN113392321A (zh) * | 2021-06-02 | 2021-09-14 | 北京三快在线科技有限公司 | 一种信息推荐方法、装置、电子设备及存储介质 |
| CN114547244A (zh) * | 2022-02-17 | 2022-05-27 | 北京沃东天骏信息技术有限公司 | 用于确定信息的方法和装置 |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114547244A (zh) | 2022-05-27 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| JP7317791B2 (ja) | エンティティ・リンキング方法、装置、機器、及び記憶媒体 | |
| US10832658B2 (en) | Quantized dialog language model for dialog systems | |
| CN112860866B (zh) | 语义检索方法、装置、设备以及存储介质 | |
| WO2023155678A1 (zh) | 用于确定信息的方法和装置 | |
| US11321534B2 (en) | Conversation space artifact generation using natural language processing, machine learning, and ontology-based techniques | |
| US11551437B2 (en) | Collaborative information extraction | |
| US20210248471A1 (en) | Method and apparatus for creating dialogue, and storage medium | |
| CN113553414B (zh) | 智能对话方法、装置、电子设备和存储介质 | |
| CN112507706B (zh) | 知识预训练模型的训练方法、装置和电子设备 | |
| CN114861889B (zh) | 深度学习模型的训练方法、目标对象检测方法和装置 | |
| US20230088445A1 (en) | Conversational recommendation method, method of training model, device and medium | |
| US20220027575A1 (en) | Method of predicting emotional style of dialogue, electronic device, and storage medium | |
| CN115688920B (zh) | 知识抽取方法、模型的训练方法、装置、设备和介质 | |
| CN114860913B (zh) | 智能问答系统构建方法、问答处理方法及装置 | |
| CN110019849B (zh) | 一种基于注意力机制的视频关注时刻检索方法及装置 | |
| CN115309877A (zh) | 对话生成方法、对话模型训练方法及装置 | |
| CN111241838B (zh) | 文本实体的语义关系处理方法、装置及设备 | |
| CN111813909A (zh) | 一种智能问答方法和装置 | |
| US20230008897A1 (en) | Information search method and device, electronic device, and storage medium | |
| CN113239157B (zh) | 对话模型的训练方法、装置、设备和存储介质 | |
| US20230094730A1 (en) | Model training method and method for human-machine interaction | |
| CN116541536B (zh) | 知识增强的内容生成系统、数据生成方法、设备和介质 | |
| CN110991183B (zh) | 问题的谓词确定方法、装置、设备及存储介质 | |
| CN113360683B (zh) | 训练跨模态检索模型的方法以及跨模态检索方法和装置 | |
| CN117688946A (zh) | 基于大模型的意图识别方法、装置、电子设备和存储介质 |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| 121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 23755682 Country of ref document: EP Kind code of ref document: A1 |
|
| WWE | Wipo information: entry into national phase |
Ref document number: 2023755682 Country of ref document: EP |
|
| ENP | Entry into the national phase |
Ref document number: 2023755682 Country of ref document: EP Effective date: 20240917 |