Text association method and device based on combination model, equipment and storage medium

Abstract

The invention relates to the field of data processing, and provides a character association method based on a combined model, which comprises the following steps of: firstly, predicting related characters to be input by a current seat based on an association prediction model and the current characters input by the seat to obtain multiple groups of related characters; and then based on the target language model of the personal input habit of the current seat, scoring each group of predicted related characters, and screening out related characters conforming to the input habit of the current seat from each group of related characters according to scores to serve as associative characters. Therefore, through two combined models of the association prediction model and the target language model, the association characters which are related to the current characters and accord with the input habits of the current seat are predicted, the accuracy of recommending the association characters through the input method is improved, the manual input work of the seat is reduced, the operation efficiency of the seat is improved, and the user experience is improved. Therefore, the waiting time of the user is reduced, and the user experience is further improved.

Description

Technical Field

[0001] The present invention relates to the fields of data processing and financial technology, and in particular to a text association method, device, computer equipment and computer-readable storage medium based on a combination model. Background Art

[0002] As people's living standards improve, they pay more and more attention to insurance products. As the number of users purchasing insurance products increases, the workload of insurance product agents also increases. Currently, insurance product agents need to type manually to communicate with users, but the accuracy of the associated text recommended by the current input method's associated input function is low, and the probability of being adopted by the agent is low. Therefore, the insurance product agent needs to manually enter all the sentences to complete the reply operation of the user's relevant information, which leads to the low efficiency of the current insurance product agent's operation.

[0003] Therefore, how to solve the low accuracy of the current input method's recommended associative text has become a technical problem that needs to be solved urgently. Summary of the invention

[0004] The main purpose of the present invention is to provide a method, device, equipment and computer-readable storage medium for character association based on a combination model, aiming to solve the technical problem of low accuracy of the current input method in recommending associated characters.

[0005] To achieve the above object, the present invention provides a text association method based on a combination model, the text association method comprising the following steps:

[0006] When an agent input operation is detected, the current text input by the current agent is obtained;

[0007] Based on a preset associative prediction model, predict the related words of the current word to obtain at least one group of the related words;

[0008] When there are multiple groups of related words, scoring each group of related words based on the target language model;

[0009] Based on the scores of each group of related characters, the associated characters corresponding to the current characters are determined, and the associated characters are displayed on the current input interface of the current seat.

[0010] In addition, to achieve the above-mentioned purpose, the present invention further provides a text association device based on a combination model, and the text association device based on a combination model comprises:

[0011] The seat input monitoring module is used to obtain the current text input by the current seat when the seat input operation is detected;

[0012] A related text prediction module, used for predicting related texts of the current text based on a preset associative prediction model to obtain at least one group of related texts;

[0013] A related text scoring module, used for scoring each group of related texts based on the target language model when there are multiple groups of related texts;

[0014] Based on the scores of each group of related characters, the associated characters corresponding to the current characters are determined, and the associated characters are displayed on the current input interface of the current seat.

[0015] In addition, to achieve the above-mentioned purpose, the present invention also provides a computer device, which includes a processor, a memory, and a text association program based on a combination model stored in the memory and executable by the processor, wherein when the text association program based on the combination model is executed by the processor, the steps of the text association method based on the combination model as described above are implemented.

[0016] In addition, to achieve the above-mentioned purpose, the present invention also provides a computer-readable storage medium, on which a text association program based on a combination model is stored, wherein when the text association program based on a combination model is executed by a processor, the steps of the text association method based on a combination model as described above are implemented.

[0017] The present application provides a text association method based on a combination model, wherein the method obtains the current text input by the current agent when an agent input operation is detected; predicts the related text of the current text based on a preset association prediction model to obtain at least one group of the related text; when there are multiple groups of the related text, scores each group of the related text based on a target language model; determines the associated text corresponding to the current text based on the score of each group of related text, and displays the associated text on the current input interface of the current agent. In the above manner, the present application first predicts the related text that the current agent is about to input based on the association prediction model and the current text input by the agent to obtain multiple groups of related text, and then scores each group of predicted related text based on the target language model of the current agent's personal input habits, and screens out related text that meets the current agent's input habits from each group of related text according to the score as the associated text. Therefore, through the combination of the associative prediction model and the target language model, the associative text related to the current text and in line with the current agent's input habits is predicted, which improves the accuracy of the associative text recommended by the input method, reduces the manual input work of the agent, improves the agent's work efficiency, thereby reducing user waiting time and further improving user experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 It is a flow chart of the first embodiment of the character association method based on the combination model of the present invention;

[0019] Figure 2 It is a flow chart of the second embodiment of the character association method based on the combination model of the present invention;

[0020] Figure 3 It is a flow chart of a third embodiment of the character association method based on the combination model of the present invention;

[0021] Figure 4 It is a schematic diagram of a text association device based on a combination model of the present invention;

[0022] Figure 5 The figure is a schematic diagram of the hardware structure of a computer device involved in an embodiment of the present invention.

[0023] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0024] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0025] The flowcharts shown in the accompanying drawings are only examples and do not necessarily include all the contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps may also be decomposed, combined or partially merged, so the actual execution order may change according to actual conditions.

[0026] It should be understood that the terms used in this application specification are only for the purpose of describing specific embodiments and are not intended to limit the application. As used in this application specification and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an" and "the" are intended to include plural forms.

[0027] It should be further understood that the term “and / or” used in the specification and appended claims refers to any combination and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0028] The text association method based on the combination model involved in the embodiment of the present invention is mainly applied to computer equipment. The text association generation device based on the combination model can be a PC, a portable computer, a mobile terminal or other equipment with display and processing functions.

[0029] In conjunction with the accompanying drawings, some embodiments of the present application are described in detail below. In the absence of conflict, the following embodiments and features in the embodiments can be combined with each other.

[0030] Reference Figure 1 , Figure 1 It is a flow chart of the first embodiment of the text association method based on the combination model of the present invention.

[0031] like Figure 1 As shown, an embodiment of the present invention provides a text association method based on a combination model, and the text association method based on a combination model includes steps S10 to S30.

[0032] In this embodiment, the text association method based on the combination model is applied to the card number customization and review system, and the text association method based on the combination model includes the following steps:

[0033] Step S10: when an agent input operation is detected, obtaining the current text input by the current agent;

[0034] In this embodiment, the current text includes characters or words, and the input operation of the agent during the operation is monitored in real time. At the same time, the characters or words manually input by the agent are obtained, and then predicted through model combination to obtain the corresponding associated text, and the associated question is displayed on the input display interface corresponding to the agent to prompt the agent, thereby reducing the manual input operation of the agent and improving the agent's work efficiency.

[0035] Step S20: predicting related words of the current word based on a preset association prediction model to obtain at least one group of related words;

[0036] In this embodiment, the associative prediction model includes a hidden Markov model and a Viterbi model, and a hidden Markov + ALBERT model combination is used to prompt the agent to make associative input, reduce the agent's manual input, and improve the agent's work efficiency.

[0037] Among them, the Hidden Markov Model (HMM) is a statistical model used to describe a Markov process with hidden unknown parameters. The basis of HMM is the Markov chain, which is a random process that changes from one state to another in the state space. This process requires the "memoryless" property: the probability distribution of the next state can only be determined by the current state, and the events before it in the time series are irrelevant. This specific type of "memorylessness" is called the Markov property. The hidden Markov model calculates the occurrence of hidden states by analyzing visible data. And the hidden states are analyzed to estimate possible future observation patterns.

[0038] The current text is input into the trained hidden Markov model to predict the word or phrase that the current seat is about to input. The current text is used as the current state of the current seat, and the next state of the current seat and the probability of the next state are encoded through the hidden Markov model, and decoded through the Viterbi model to obtain multiple groups of related texts corresponding to the current text.

[0039] Step S30: when there are multiple groups of related words, scoring each group of related words based on the target language model;

[0040] In this embodiment, a target language model that conforms to the current agent's personal input habits is obtained from the language model. Through the target language model, each group of related words is scored according to the current agent's personal input habits.

[0041] In order to solve the problem of BERT model having too many parameters and slow training, the ALBERT model is used as the language model. The ALBERT model is an advanced version of the BERT model, which can reduce the number of parameters while improving model performance. The ALBERT model significantly reduces the number of parameters without significantly affecting model performance by factorizing Embedding parameters and sharing cross-layer parameters.

[0042] Among them, the factorization of the Embedding parameter decomposes the large word embedding matrix and decomposes the embedding matrix into two small matrices, while the cross-layer parameter sharing avoids the increase of the number of parameters as the depth of the network increases; at the same time, a self-supervised sentence prediction target SOP (sentence-order prediction) is adopted to express the focus on the coherence between sentences to solve the invalidity problem of NSP (next sentence prediction) in BERT. Therefore, the ALBERT model successfully maintained or even improved the model performance while reducing the model size through parameter reduction technology and improved pre-training tasks.

[0043] Step S40: based on the scores of each group of related characters, determine the associated characters corresponding to the current characters, and display the associated characters on the current input interface of the current seat.

[0044] In this embodiment, the higher the score, the more the group of related words conforms to the personal input habits of the current seat. Therefore, the related words with the highest score, or the top several groups of related words, can be obtained as the associated words corresponding to the current words. The associated words are displayed through the current input interface of the current seat for the current seat to select, thereby avoiding the current seat from having to input manually.

[0045] Furthermore, when there are multiple groups of related words, before scoring each group of related words based on the target language model, the method further includes:

[0046] The historical text input records of the current seat are obtained, and based on each group of characters and associated characters in the historical text input records, the initial language model is trained twice to obtain the target language model.

[0047] Furthermore, the method of determining the associated text corresponding to the current text based on the scores of each group of related texts, and after displaying the associated text on the current input interface of the current seat, further includes:

[0048] The associated text selected by the current agent is monitored, and the current text and the associated text are obtained as a group of training samples, and added to a preset sample library corresponding to the associated prediction model to update the sample library corresponding to the associated prediction model.

[0049] In this embodiment, after training the prediction model applicable to all seats based on the input text and its corresponding next state based on the big data, the historical input text and its corresponding next state in the input method of each insurance product seat (i.e., the input method account corresponding to each insurance product seat) are further used as training samples for each model, and the prediction model that conforms to the personal input habits of each insurance product seat is further trained. When the text prediction is performed based on each prediction model corresponding to each insurance product seat in the future, the input text and its corresponding associated text (i.e., the relevant text adopted by the insurance product seat) are further used as subsequent training samples to update the sample library, thereby improving the prediction accuracy of the prediction model.

[0050] This embodiment provides a text association method based on a combination model, which obtains the current text input by the current agent when an agent input operation is detected; predicts the related text of the current text based on a preset association prediction model to obtain at least one group of the related text; when there are multiple groups of the related text, scores each group of the related text based on a target language model; determines the associated text corresponding to the current text based on the score of each group of related text, and displays the associated text on the current input interface of the current agent. In the above manner, the present application first predicts the related text that the current agent is about to input based on the association prediction model and the current text input by the agent to obtain multiple groups of related text, and then scores each group of predicted related text based on the target language model of the current agent's personal input habits, and screens out related text that meets the current agent's input habits from each group of related text according to the score as the associated text. Therefore, through the combination of the associative prediction model and the target language model, the associative text related to the current text and in line with the current agent's input habits is predicted, which improves the accuracy of the associative text recommended by the input method, reduces the manual input work of the agent, improves the agent's work efficiency, thereby reducing user waiting time and further improving user experience.

[0051] Reference Figure 2 , Figure 2 It is a flow chart of the second embodiment of the text association method based on the combination model of the present invention.

[0052] Based on the above Figure 1 In the embodiment shown, in this embodiment, the step S20 specifically includes:

[0053] Step S21, predicting the related characters of the current character based on the hidden Markov model to obtain at least one set of codes corresponding to the related characters;

[0054] Step S22, based on the Viterbi model, decoding the codes corresponding to each group of related characters to obtain each group of related characters.

[0055] Furthermore, before predicting the related characters of the current character based on the hidden Markov model to obtain at least one set of codes corresponding to the related characters, the method further includes:

[0056] Based on the preset observation sequence samples and the maximum expectation algorithm EM algorithm, the initial parameters of the hidden Markov model are obtained;

[0057] Based on the forward probability, the backward probability and the known observation sequence samples, the hidden Markov model is iteratively trained until the parameters of the hidden Markov model converge, the target parameters of the hidden Markov model are obtained, and the training of the hidden Markov model is completed.

[0058] In this embodiment, training the hidden Markov model is to train the parameters of the hidden Markov model. T}, it is known that the observation sequence corresponding to the sample sequence to be predicted is O = {o1, o2, …, o T}, it is necessary to estimate the initial parameters of the hidden Markov model λ = (π, A, B), so that under this parameter, the probability value of the hidden Markov model's prediction result for the predicted sample sequence is the observed sequence is maximized.

[0059] It is necessary to use the Expectation-Maximization algorithm (EM) to learn the parameters of the hidden Markov model.

[0060] The EM algorithm introduces hidden variables to perform maximum likelihood estimation or maximum a posteriori probability estimation on the parameters of a probability parameter model (i.e., a hidden Markov model).

[0061] Specifically, for m independent training samples x=(x (1) ,x (2) ,…,x (m) ), the corresponding implicit variable data z=(z (1) ,z (2) ,…,z (m) ), the model parameter of the sample is θ, and the observed data x (1) The probability of P(x (1) |θ) Assuming that the hidden variable is z, the final parameter can be obtained by continuously adjusting the model parameter θ and maximizing the logarithmic function. The calculation formula of the logarithmic function of θ is:

[0062]

[0063] After adding hidden variables, the objective function is adjusted to determine the appropriate θ and z so that the logarithmic function is maximized. The logarithmic function calculation formula of θ and z is:

[0064]

[0065] In a specific embodiment, the training process based on the EM algorithm is as follows:

[0066] Input the training sample x=(x (1) ,x (2) ,…,x(m) ), joint distribution P(x,z|θ), conditional distribution P(z|x,θ), where z is a hidden variable, m is the maximum number of iterations, and the initial value of the parameter θ of the initialization model is θ 0 , enter the EM loop iteration from step 1 to step m, and finally output the model parameter θ. Among them, the expectation of the conditional probability of the hidden variable is calculated:

[0067]

[0068] Q i (z (i) )=P(z (i) |x (i) ,θ)

[0069] Maximize L(θ) to get θ, where the formula for calculating θ combined with the Q function is as follows:

[0070]

[0071] The process of calculating and training Markov parameters based on the EM algorithm is as follows:

[0072] First, construct the Q function at the kth step of the iteration as:

[0073] Q k (λ)=Σ I P(I,O|λ k )lnP(I,O|λ)

[0074] Among them, O is the observed variable and I is the corresponding hidden variable.

[0075] At step k, the parameter λ is known (k) , P(I,O|λ (k) ) is expanded using the conditional probability formula to obtain:

[0076]

[0077] Given P(I,O|λ (k) ), and for the parameter P(I,O|λ) whose position is to be solved, it is also expanded:

[0078]

[0079] Combine Ⅰ, Ⅱ and Ⅲ to optimize P(I,O|λ), where Ⅰ, Ⅱ and Ⅲ are:

[0080]

[0081] Then, the parameters are solved based on the forward probability and backward probability. When the observation sequence corresponding to the known prediction sample sequence is O = {o1, o2, ..., oT}, when the model parameters λ=(π,A,B) are known, the probability of occurrence of the observation sequence P(O|λ) can be solved by three methods, namely, exhaustive method, forward algorithm and backward algorithm. Among them, the exhaustive method consumes relatively high time complexity and space complexity. The forward algorithm and backward algorithm introduce forward probability and backward probability respectively to dynamically program and solve P(O|λ).

[0082] The solution process based on forward probability is: given the hidden Markov model parameter λ, the state at a certain time t is q i , and the observation sequence is o1,o2,…,o t The joint probability distribution of t (i), α t The calculation formula of (i) is as follows:

[0083] α t (i)=P(o1,o2,…,o t ,i t =q i |λ),i=1,2,…,N

[0084] The solution process based on backward probability is: given a hidden Markov model λ, at a certain time t, in state q i Under the condition of t+1 ,o t+2 ,…,o T The conditional probability of t (i), β t The calculation formula of (i) is as follows:

[0085] β t (i) = P(o t+1 ,o t+2 ,…,o T |i t =q i ,λ),i=1,2,…,N

[0086] Define the state q at time t i The probability is γ t (i), γ t The calculation formula of (i) is as follows:

[0087]

[0088] Given the model parameters λ and observation O, at time t, in state q i And at time t+1 it is in state q j The probability ξ t (i,j),ξ tThe calculation formula for (i,j) is as follows:

[0089]

[0090] Among them, the parameter π i ,α ij ,b j The update formula of (k) is:

[0091] π i =γ1(i)

[0092]

[0093] Get the parameter value λ for step k+1 (k+1) , and then iterate until the parameters converge.

[0094] After predicting the related characters of the current character based on the hidden Markov model and obtaining at least one group of codes corresponding to the related characters, the codes corresponding to each group of the related characters are further decoded by the Vibit model.

[0095] Given the model parameters λ = (π, A, B) and the observation sequence O, find the sequence state I = {i1, i2, …, i T}, the model parameters have been obtained by the above EM algorithm training, the goal is to solve P(I|O), the calculation formula of the maximum P(I|O) is as follows:

[0096]

[0097] In a hidden Markov chain, there are N values ​​corresponding to the state at any time t. If the state sequence is expanded by value, a Lattice network can be obtained, and then the Viterbi algorithm is used for decoding.

[0098] Among them, i = 1, 2, ..., N, fixed at the state selection at time t, let i t =i, that is, to find the shortest path from this point to the starting point (the path with the highest probability), the formula is as follows:

[0099]

[0100] The recursive formula is as follows:

[0101]

[0102] We further define all single paths (i1,…,i t-1 ,i) The calculation of the t-1th node of the path with the highest probability is as follows:

[0103]

[0104] The processing steps of the Viterbi algorithm are as follows:

[0105] 1. First, initialize the time, that is, t = 1:

[0106]

[0107] ψ1(i)=0

[0108] 2. Substitute into the recursive formula (t=2,3,…,T)

[0109]

[0110] 3. Stop calculation and get the result:

[0111]

[0112] 4. Backtracking of the optimal path (t = T-1, ..., 2, 1)

[0113]

[0114] Thus, the next state text corresponding to the current text is obtained as a group of related texts, and each group of related texts includes at least one character or word.

[0115] Reference Figure 3 , Figure 3 This is a flow chart of the third embodiment of the text association method based on the combination model of the present invention. Figure 1 In the embodiment shown, in this embodiment, the step S30 specifically includes:

[0116] Step S31, when there are multiple groups of related characters, obtaining each character and / or each word in each group of related characters;

[0117] Step S32, scoring each group of related texts based on the Albert bidirectional language model and the context of each character and / or each word in each group of related texts to obtain the score of each group of related texts.

[0118] In this embodiment, given a single word or phrase, Hidden Markov can decode several candidate sentences. Assuming that one of the sentences is ω=(ω1,ω2,…,ω1), the score of the sentence can be calculated by the Albert bidirectional language model. Each word is no longer determined by the previous or next context, but by the context together, making the language information feature extraction richer. The calculation formula is as follows:

[0119]

[0120] Finally, according to the scoring results, a preset number of association prompts with high scores are selected for association prompts.

[0121] Through the above-mentioned methods, this embodiment solves the time-consuming, labor-intensive and inefficient problems of current agents in manual input scenario operations. By adopting the Hidden Markov + ALBERT model combination, the ALBERT model improves model performance while reducing the model size through parameter reduction technology and improved pre-training tasks. And by real-time monitoring of the agent's input operations during the operation process, the characters or words manually input by the agent are obtained at the same time, and then the corresponding associative input is obtained through model combination training and the agent is prompted, thereby reducing the manual input operations of the agent and improving the agent's operating efficiency.

[0122] Furthermore, before determining the associated character corresponding to the current character based on the scores of each group of related characters, the method further includes:

[0123] The interface display preference corresponding to the current seat is obtained, and the maximum number of display associated characters of the current input interface is determined based on the interface display preference, so as to determine the number of the associated characters to be displayed based on the maximum number of display associated characters.

[0124] In this embodiment, based on the input interface display preference of the current agent, the maximum number of associated characters displayed on the current input interface is determined based on the interface display preference, so as to determine the number of associative characters to be displayed based on the maximum number of associated characters displayed, such as the number of associative characters displayed (e.g., 3 groups are initially displayed, and the agent can increase the probability of associative characters being adopted by increasing the number of display groups), the display size of associative characters, and the display form of associative characters.

[0125] See also Figure 4 , Figure 4 The present invention provides a functional module diagram of a text association device based on a combination model according to an embodiment of the present application.

[0126] like Figure 4 As shown, the text association device 400 based on the combination model includes:

[0127] The seat input monitoring module 10 is used to obtain the current text input by the current seat when the seat input operation is detected;

[0128] A related text prediction module 20, configured to predict related texts of the current text based on a preset associative prediction model, and obtain at least one group of related texts;

[0129] A related text scoring module 30, for scoring each group of related texts based on a target language model when there are multiple groups of related texts;

[0130] The associative text display module 40 is used to determine the associative text corresponding to the current text based on the scores of each group of related texts, and display the associative text on the current input interface of the current seat.

[0131] Furthermore, the associative text display module 40 specifically includes:

[0132] A text input monitoring unit, used to obtain the current text input by the current agent when an agent input operation is detected;

[0133] A related text acquisition unit, configured to predict related texts of the current text based on a preset association prediction model, and obtain at least one group of related texts;

[0134] A related text scoring unit, used for scoring each group of related texts based on the target language model when there are multiple groups of related texts;

[0135] The associative text display unit is used to determine the associative text corresponding to the current text based on the scores of each group of related texts, and display the associative text on the current input interface of the current seat.

[0136] Furthermore, the related text prediction module 20 specifically includes:

[0137] A related text encoding unit, used for predicting related texts of the current text based on the hidden Markov model, and obtaining at least one set of codes corresponding to the related texts;

[0138] The related text decoding unit is used to decode the codes corresponding to each group of related texts based on the Viterbi model to obtain each group of related texts.

[0139] Furthermore, the text association device 400 based on the combination model further includes:

[0140] An initial parameter determination module is used to obtain the initial parameters of the hidden Markov model based on a preset observation sequence sample and a maximum expectation algorithm (EM algorithm);

[0141] The model iteration training module is used to iteratively train the hidden Markov model based on the forward probability, the backward probability and the known observation sequence samples until the parameters of the hidden Markov model converge, obtain the target parameters of the hidden Markov model, and complete the training of the hidden Markov model.

[0142] Furthermore, the related text scoring module 30 includes:

[0143] A word acquisition unit, used for acquiring each word and / or each phrase in each group of related characters when there are multiple groups of related characters;

[0144] The word scoring unit is used to score each group of related words based on the Albert bidirectional language model and the context of each word and / or each term in each group of related words to obtain the score of each group of related words.

[0145] Furthermore, the text association device 400 based on the combination model further includes:

[0146] The associative text determination module is used to obtain the interface display preference corresponding to the current seat, and determine the maximum number of display associated texts of the current input interface based on the interface display preference, so as to determine the number of the associative texts to be displayed based on the maximum number of display associated texts.

[0147] Furthermore, the text association device 400 based on the combination model further includes:

[0148] The training sample updating module is used to monitor the associative text selected by the current agent, and obtain the current text and the associated text as a group of training samples, and add them to the preset sample library corresponding to the associative prediction model to update the sample library corresponding to the associative prediction model.

[0149] Furthermore, the text association device 400 based on the combination model further includes:

[0150] The language model training module is used to obtain the historical text input records of the current seat, and perform secondary training on the initial language model based on each group of text and its associated text in the historical text input records to obtain the target language model.

[0151] It should be noted that those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the above-described device and each module can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here.

[0152] The above-mentioned device can be implemented in the form of a computer program. Figure 5 Runs on the computer device shown.

[0153] See also Figure 5 , Figure 5 1 is a schematic block diagram of a computer device provided in an embodiment of the present application. The computer device may be a server.

[0154] See also Figure 5The computer device includes a processor, a memory and a network interface connected through a system bus, wherein the memory may include a non-volatile storage medium and an internal memory.

[0155] The non-volatile storage medium can store an operating system and a computer program. The computer program includes program instructions, and when the program instructions are executed, the processor can execute any one of the text association methods based on the combination model.

[0156] The processor is used to provide computing and control capabilities and support the operation of the entire computer equipment.

[0157] The internal memory provides an environment for the operation of the computer program in the non-volatile storage medium. When the computer program is executed by the processor, the processor can execute any text association method based on the combination model.

[0158] The network interface is used for network communication, such as sending assigned tasks, etc. Those skilled in the art will understand that Figure 5 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0159] It should be understood that the processor may be a central processing unit (CPU), and the processor may also be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among them, the general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0160] In one embodiment, the processor is used to run a computer program stored in the memory to implement the following steps:

[0161] When an agent input operation is detected, the current text input by the current agent is obtained;

[0162] Based on a preset associative prediction model, predict the related words of the current word to obtain at least one group of the related words;

[0163] When there are multiple groups of related words, scoring each group of related words based on the target language model;

[0164] Based on the scores of each group of related characters, the associated characters corresponding to the current characters are determined, and the associated characters are displayed on the current input interface of the current seat.

[0165] In one embodiment, the processor is used to run a computer program stored in the memory to implement the following steps:

[0166] Based on the hidden Markov model, predict the related characters of the current character to obtain at least one set of codes corresponding to the related characters;

[0167] Based on the Viterbi model, the codes corresponding to each group of related characters are decoded to obtain each group of related characters.

[0168] In one embodiment, the processor is used to run a computer program stored in the memory to implement the following steps:

[0169] Based on the preset observation sequence samples and the maximum expectation algorithm EM algorithm, the initial parameters of the hidden Markov model are obtained;

[0170] Based on the forward probability, the backward probability and the known observation sequence samples, the hidden Markov model is iteratively trained until the parameters of the hidden Markov model converge, the target parameters of the hidden Markov model are obtained, and the training of the hidden Markov model is completed.

[0171] In one embodiment, the processor is used to run a computer program stored in the memory to implement the following steps:

[0172] When there are multiple groups of related characters, obtaining each character and / or each word in each group of related characters;

[0173] Based on the Albert bidirectional language model and the context of each character and / or each word in each group of related texts, each group of related texts is scored to obtain the score of each group of related texts.

[0174] In one embodiment, the processor is used to run a computer program stored in the memory to implement the following steps:

[0175] The interface display preference corresponding to the current seat is obtained, and the maximum number of display associated characters of the current input interface is determined based on the interface display preference, so as to determine the number of the associated characters to be displayed based on the maximum number of display associated characters.

[0176] In one embodiment, the processor is used to run a computer program stored in the memory to implement the following steps:

[0177] The associated text selected by the current agent is monitored, and the current text and the associated text are obtained as a group of training samples, and added to a preset sample library corresponding to the associated prediction model to update the sample library corresponding to the associated prediction model.

[0178] In one embodiment, the processor is used to run a computer program stored in the memory to implement the following steps:

[0179] The historical text input records of the current seat are obtained, and based on each group of characters and associated characters in the historical text input records, the initial language model is trained twice to obtain the target language model.

[0180] A computer-readable storage medium is also provided in an embodiment of the present application, wherein the computer-readable storage medium stores a computer program, wherein the computer program includes program instructions, and the processor executes the program instructions to implement any one of the combination model-based text association methods provided in the embodiments of the present application.

[0181] The computer-readable storage medium may be an internal storage unit of the computer device described in the foregoing embodiment, such as a hard disk or memory of the computer device. The computer-readable storage medium may also be an external storage device of the computer device, such as a plug-in hard disk, a smart memory card (Smart Medi a Card, SMC), a secure digital (Secure Digital, SD) card, a flash memory card (Flash Card), etc., equipped on the computer device.

[0182] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of various equivalent modifications or replacements within the technical scope disclosed in the present application, and these modifications or replacements should be included in the protection scope of the present application. Therefore, the protection scope of the present application shall be based on the protection scope of the claims.

Claims

1. A text association method based on a combination model, characterized in that: The word association method comprises the following steps: When an agent input operation is detected, the current text input by the current agent is obtained; Based on a preset associative prediction model, predict the related words of the current word to obtain at least one group of the related words; When there are multiple groups of related words, scoring each group of related words based on the target language model; Based on the scores of each group of related characters, the associated characters corresponding to the current characters are determined, and the associated characters are displayed on the current input interface of the current seat.

2. The text association method based on the combination model as claimed in claim 1, characterized in that: The association prediction model includes a hidden Markov model and a Viterbi model. Based on the preset association prediction model, the related words of the current word are predicted to obtain at least one group of the related words, including: Based on the hidden Markov model, predict the related characters of the current character to obtain at least one set of codes corresponding to the related characters; Based on the Viterbi model, the codes corresponding to each group of related characters are decoded to obtain each group of related characters.

3. The text association method based on the combination model as claimed in claim 2, characterized in that: Before predicting the related characters of the current character based on the hidden Markov model to obtain at least one set of codes corresponding to the related characters, the method further includes: Based on the preset observation sequence samples and the maximum expectation algorithm EM algorithm, the initial parameters of the hidden Markov model are obtained; Based on the forward probability, the backward probability and the known observation sequence samples, the hidden Markov model is iteratively trained until the parameters of the hidden Markov model converge, the target parameters of the hidden Markov model are obtained, and the training of the hidden Markov model is completed.

4. The text association method based on the combination model as claimed in claim 1, characterized in that: The target language model is an Albert bidirectional language model. When there are multiple groups of related texts, scoring each group of related texts based on the target language model includes: When there are multiple groups of related characters, obtaining each character and / or each word in each group of related characters; Based on the Albert bidirectional language model and the context of each character and / or each word in each group of related texts, each group of related texts is scored to obtain the score of each group of related texts.

5. The text association method based on the combination model as claimed in claim 1, characterized in that: Before determining the associated character corresponding to the current character based on the scores of each group of related characters, the method further includes: The interface display preference corresponding to the current seat is obtained, and the maximum number of display associated characters of the current input interface is determined based on the interface display preference, so as to determine the number of the associated characters to be displayed based on the maximum number of display associated characters.

6. The text association method based on the combination model as claimed in claim 1, characterized in that: The method further includes determining the associated text corresponding to the current text based on the scores of each group of related texts, and after displaying the associated text on the current input interface of the current seat: The associated text selected by the current agent is monitored, and the current text and the associated text are obtained as a group of training samples, and added to a preset sample library corresponding to the associated prediction model to update the sample library corresponding to the associated prediction model.

7. The text association method based on the combination model according to any one of claims 1 to 6, characterized in that: When there are multiple groups of related words, before scoring each group of related words based on the target language model, the method further includes: The historical text input records of the current seat are obtained, and based on each group of characters and associated characters in the historical text input records, the initial language model is trained twice to obtain the target language model.

8. A text association device based on a combination model, characterized in that: The text association device based on the combination model includes: The seat input monitoring module is used to obtain the current text input by the current seat when the seat input operation is detected; A related text prediction module, used for predicting related texts of the current text based on a preset associative prediction model to obtain at least one group of related texts; A related text scoring module, used for scoring each group of related texts based on the target language model when there are multiple groups of related texts; The associative text display module is used to determine the associative text corresponding to the current text based on the scores of each group of related texts, and display the associative text on the current input interface of the current seat.

9. A computer device, characterized in that: The computer device includes a processor, a memory, and a text association program based on a combination model stored in the memory and executable by the processor, wherein the text association program based on a combination model, when executed by the processor, implements the steps of the text association method based on a combination model as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a text association program based on a combination model, wherein when the text association program based on a combination model is executed by a processor, the steps of the text association method based on a combination model as described in any one of claims 1 to 7 are implemented.

Images