Artificial intelligence dialogue system construction method and device based on c-transform, and medium

Abstract

The application discloses a kind of based on C-transform artificial intelligence conversation system construction method, device and medium, method includes: to input position embedding value sequence is calculated to obtain two-dimensional conjugate sequence, then use the method of random segmentation to generate alternative input two-dimensional data sequence, then through C-transform whether alternative input two-dimensional data sequence has conjugacy is judged, to alternative input two-dimensional data sequence is short sentence exchange pairing, obtains paired data sequence, whether paired data sequence is judged to converge, the probability and quantity of output short sentence sequence are calculated, the maximum short sentence of probability is output, all output short sentence is linked together as output sequence, compare optimization and obtain the parameter value of error minimization artificial intelligence conversation system of embedding numerical sequence to output sequence. Based on the conversation output result of C-transform artificial intelligence conversation system construction method is more accurate, sentence analysis efficiency is high.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of artificial intelligence software, in particular to a C-transform-based artificial intelligence dialogue system construction method, device, terminal and medium. BACKGROUND

[0002] The current artificial intelligence dialogue system, especially the system based on neural network algorithm, including ChatGPT, is data-driven, and the law of numerical calculation is found from a large number of labeled or learned data. People are difficult to know why it works, especially the so-called "black box" model, except for input and output, those hidden layers, neurons, activation functions, self-attention mechanisms or correlation analysis are difficult to explain, and a large amount of data is needed to learn, in addition, the model is also very complex, such as ChatGPT, which produces decentralized results while consuming a large amount of memory to save basic code information through complex operations. Another problem is that it is difficult to establish a relationship between two words that are far apart, and the solution is very wasteful of memory, such as: ChatGPT has a copy of information for each position. Another difficulty is the calculation of the correlation between two vectors. The problem with traditional dot multiplication is that a good normalization operation is required, which cannot be obtained until the entire data set is understood. After the entire data set changes, the normalization operation needs to be adjusted. Finally, in ChatGPT, the character encoding is fixed, similar to one-hot (one-hot encoding), which is actually using an N-bit status register to encode N states, each state has an independent register bit, and only one bit in these register bits is valid, which means that only one state is allowed. In addition to the difference between each encoding, this encoding cannot reflect the relationship between words and sentences, that is, Query is independent of the information of the sentence and is fixed. SUMMARY

[0003] In view of the defects in the prior art, the C-transform-based artificial intelligence dialogue system construction method, device, terminal and medium provided by the present application have more accurate dialogue output results and high sentence analysis efficiency.

[0004] In the first aspect, the C-transform-based artificial intelligence dialogue system construction method provided by the present application comprises the following steps:

[0005] S1: Obtain demonstration data of an artificial intelligence dialogue system, map each word to a corresponding code value using the question part of the demonstration data, and obtain an input position embedding value sequence in combination with the positions of each word;

[0006] S2: Calculate the input position embedding value sequence by the Talmud algorithm to obtain a two-dimensional conjugate sequence;

[0007] S3: obtain random data, segment the random data to obtain segmented random input data, fuse the segmented random input data and the two-dimensional conjugate sequence through C-transformation, and generate a candidate input two-dimensional data sequence;

[0008] S4: determine whether the candidate input two-dimensional data sequence has conjugacy through C-transformation, if yes, execute step S5, and if no, return to step S3;

[0009] S5: perform short sentence exchange pairing on the candidate input two-dimensional data sequence to obtain a paired data sequence, determine whether the paired data sequence converges, if not, return to step S3, and if yes, execute step S6;

[0010] S6: randomly segment the reply part in the demonstration data to obtain an embedding numerical sequence of each segment output short sentence;

[0011] S7: calculate the number of corresponding output short sentence sequences and the corresponding probability of all output short sentence sequences in the output by using the noise data used in the C-transformation process, the word codes of all words as learning parameters, and the paired data sequence, output the short sentence with the maximum probability, perform deep learning on all previous part output short sentences and input short sentences in turn to generate the output short sentence of the subsequent part, link all the output short sentence sequences with the maximum probability together to form an output sequence, and compare and optimize the output sequence and the embedding numerical sequence to obtain the parameter value of the artificial intelligence dialogue system with the minimum error.

[0012] In a second aspect, the application provides a C-transformation-based artificial intelligence dialogue system construction device, comprising an acquisition module, a calculation module, a fusion module, a judgment module, an analysis module, a segmentation module, and a processing module, wherein,

[0013] The acquisition module is configured to acquire demonstration data of an artificial intelligence dialogue system, map each word and the corresponding word code value by using the question part of the demonstration data, and obtain an input position embedding value sequence in combination with the positions of the words.

[0014] The calculation module is configured to calculate the input position embedding value sequence through a Tamude algorithm to obtain a two-dimensional conjugate sequence.

[0015] The fusion module is configured to obtain random data, segment the random data to obtain segmented random input data, fuse the segmented random input data and the two-dimensional conjugate sequence through C-transformation, and generate a candidate input two-dimensional data sequence.

[0016] The judgment module judges whether the candidate input two-dimensional data sequence has conjugacy through C-transformation, if having conjugacy, analyzes by using the analysis module, if not having conjugacy, fuses by using the fusion module;

[0017] The analysis module exchanges and pairs the candidate input two-dimensional data sequence to obtain a paired data sequence, judges whether the paired data sequence converges, if converging, carries out random segmentation processing by using the segmentation module, if not converging, fuses by using the fusion module;

[0018] The segmentation module carries out random segmentation on the reply part in the demonstration data to obtain an embedded numerical sequence of each segment output short sentence;

[0019] The processing module calculates the number of corresponding output short sentence sequences and the corresponding probability of all output short sentence sequences in the output by using the noise-added data used in the C-transformation process, all word codes of words as learning parameters and the paired data sequence, outputs the short sentence with the maximum probability, carries out deep learning on all previous part output short sentences and input short sentences in turn to generate the output short sentence of the following part, compares the output sequence formed by connecting all output short sentence sequences together, compares the output sequence with the embedded numerical sequence to obtain the parameter value of the artificial intelligence conversation system with the minimum error.

[0020] In a third aspect, the embodiments of the present application provide an intelligent terminal, which comprises a processor, an input device, an output device and a memory, the processor, the input device, the output device and the memory are connected with each other, the memory is used for storing a computer program, the computer program comprises program instructions, the processor is configured to invoke the program instructions and execute the method described in the above embodiments.

[0021] In a fourth aspect, the embodiments of the present application further provide a computer readable storage medium, which stores a computer program, the computer program comprises program instructions, and the program instructions make the processor execute the method described in the above embodiments when the processor executes the program instructions.

[0022] The present application has the following beneficial effects:

[0023] The embodiment of the application provides a kind of artificial intelligence dialogue system construction method, device, terminal and medium based on C-transform, by learning the code value of Chinese character, automatic segmentation, the relationship between corresponding phrase can be established, the efficiency of Chinese sentence analysis can be improved.C-transform reconstruction code information is added with noise, new output short sentence is constantly generated using the relationship between input short sentence and output short sentence of depth learning, the output sequence is obtained by connecting all the most probable output short sentence, the embedding numerical sequence obtained by comparing and optimizing output sequence and demonstration sentence segmentation is obtained, to obtain the final dialogue output result, the dialogue output result obtained is more accurate, and the sentence analysis efficiency is high. BRIEF DESCRIPTION OF DRAWINGS

[0024] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the description of the embodiments or the prior art. In all the drawings, similar elements or parts are generally identified by similar reference signs. In the drawings, each element or part is not necessarily drawn according to the actual proportion.

[0025] Figure 1 A flowchart of a C-transform-based artificial intelligence dialogue system construction method provided by the first embodiment of the present application is shown;

[0026] Figure 2 A structure diagram of a C-transform-based artificial intelligence dialogue system construction device provided by another embodiment of the present application is shown;

[0027] Figure 3 A structure diagram of an intelligent terminal provided by another embodiment of the present application is shown. DETAILED DESCRIPTION

[0028] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0029] It should be understood that, when used in the present specification and the appended claims, the terms "comprise" and "include" indicate the existence of described features, integers, steps, operations, elements, and / or components, but do not exclude one or more other features, integers, steps, operations, elements, components, and / or sets thereof.

[0030] It should also be understood that the terms used in the present application specification are for the purpose of describing particular embodiments and do not intend to limit the present application. As used in the present application specification and the appended claims, the singular forms "a," "an," and "the" are intended to include the plural forms unless the context clearly indicates otherwise.

[0031] It should be further understood that the term "and / or" used in the present application specification and the appended claims means one or more of the associated listed items as well as all possible combinations of the items and includes these combinations.

[0032] As used in the present specification and the appended claims, the term "if" can be interpreted as meaning "when" or "once" or "in response to a determination" or "in response to detecting" depending on the context. Similarly, the phrase "if it is determined" or "if [a described condition or event] is detected" can be interpreted to mean "once it is determined" or "in response to a determination" or "once [a described condition or event] is detected" or "in response to detecting [a described condition or event]" depending on the context.

[0033] It should be noted that, unless otherwise specified, the technical terms or scientific terms used in the present application should be understood as the general meaning understood by the skilled person in the field to which the present application belongs.

[0034] As Figure 1 shown, a C-transform-based artificial intelligence dialogue system construction method provided by a first embodiment of the present application is shown, which includes the following steps:

[0035] S1: Obtain demonstration data of an artificial intelligence dialogue system, map each character and the corresponding character code value using the question part of the demonstration data, and obtain an input position embedding value sequence in combination with the position of each character;

[0036] S2: Calculate the input position embedding value sequence by the Talmud algorithm to obtain a two-dimensional conjugate sequence, and this step produces a deterministic result of data;

[0037] S3: Obtain random data and perform segmentation processing on the random data to obtain segmented random input data, fuse the segmented random input data and the two-dimensional conjugate sequence through C-transform to generate an alternative input two-dimensional data sequence, and this step produces a random result of data;

[0038] S4: Determine whether the alternative input two-dimensional data sequence has conjugacy through C-transform, if yes, execute step S5, and if no, return to step S3;

[0039] S5: performing short sentence exchange pairing on the alternative input two-dimensional data sequence to obtain a paired data sequence, judging whether the paired data sequence converges, if not, returning to step S3, if yes, performing step S6;

[0040] S6: performing random segmentation on the reply part in the demonstration data to obtain an embedding numerical sequence of each segment output short sentence;

[0041] S7: using the noise-added data used in the C-transformation process, taking all word codes as learning parameters and the paired data sequence to calculate the number of corresponding output short sentence sequences and the probability of all output short sentence sequences in the output, outputting the short sentence with the maximum probability, sequentially performing deep learning on all previous part output short sentences and input short sentences to generate the output short sentences of the following part, comparing the output sequence formed by connecting all the output short sentence sequences with the maximum probability to obtain the parameter value of the artificial intelligence conversation system with the minimum error through comparison and optimization of the embedding numerical sequence.

[0042] Specifically, demonstration data of an artificial intelligence conversation system is obtained, and encoding of words and noise-added signals are fused to generate a large amount of data satisfying maximum information entropy. The question part and the reply part of the demonstration data are used to map each word and the corresponding word code value, and the input position embedding value sequence is obtained in combination with the position of each word. The input position embedding value sequence is calculated through the Tamde algorithm to obtain a two-dimensional conjugate sequence, and a deterministic result of the data is generated. The Tamde algorithm performs a fair distribution with the horizontal axis as the distribution relationship and the vertical axis as the total number. The Tamde algorithm is reversed with the vertical axis as the distribution relationship and the horizontal axis as the total number. The two are alternately rotated, and finally a stable two-dimensional conjugate sequence is obtained. Random data is used to represent all data in MCMC sampling. The segmented random input data is obtained by optimizing the segmentation of the random data. The segmented random input data and the two-dimensional conjugate sequence are fused through C-transformation to generate an alternative input two-dimensional data sequence, and a random result of the data is generated. The segmented random data and the two-dimensional conjugate sequence are rejected and accepted or combined with other two processes to generate an alternative input two-dimensional data sequence. Through the deterministic result and the random result of the data, an optimized result can be obtained.

[0043] The C-transformation includes input data and noise, and the specific method of the C-transformation includes:

[0044] The data variable is assigned the input data;

[0045] The noise-added data is set, the initial value of the noise amplitude is set, and the step variable is set to zero;

[0046] The noise-added data is multiplied by the initial value of the amplitude and added to the data variable;

[0047] normalizing the data variable;

[0048] determining whether the normalized data variable is white noise, if yes, returning the step variable as output, if not, automatically adding 1 to the step variable on the basis of the original step variable, adjusting the noise amplitude, and returning to execute the operation of multiplying the noise-added data by the initial value of the amplitude to add to the data variable, and calculating the step number needed to generate white noise. The value of C-change or the step number can be verified by calculation to determine whether it is close to the entropy value, and the degree of disorder can be obtained.

[0049] The process of C-change here can be regarded as a process based on differential operation. According to Taylor expansion, the original code information can be reconstructed by using each order derivative, and the dispersion result can be obtained by using multi-order C-change. The problem of establishing a relationship between two words far apart can be solved by using noise-added C-change, which can establish a long-term relationship and save memory. The method is to reconstruct the code information by using optimized C-change of each order through a method similar to Taylor expansion. The larger the value or step number of C-change, the more difficult it is for the data to become white noise, which means that the data is more regular. Therefore, C-change and the degree of disorder are inversely proportional.

[0050] The specific method of fusing the segmented random data and the two-dimensional conjugate sequence through C-change includes:

[0051] aligning the segmented random data and the two-dimensional conjugate sequence to obtain aligned random data and aligned two-dimensional conjugate sequence.

[0052] Specifically, the 1st to Nth order C-change values are calculated by using the segmented random data and the first noise to obtain the aligned random data, and N is the alignment length.

[0053] The 1st to Nth order C-change values are calculated by using the two-dimensional conjugate sequence and the second noise to obtain the aligned two-dimensional conjugate sequence.

[0054] The aligned random data and the aligned two-dimensional conjugate sequence are fused.

[0055] Specifically, the specific method of fusing the aligned random data and the aligned two-dimensional conjugate sequence includes:

[0056] The first C-change value is calculated by using the aligned random data and the aligned two-dimensional conjugate sequence for noise addition.

[0057] The second C-change value is calculated by using the aligned two-dimensional conjugate sequence and the aligned random data for noise addition.

[0058] The 1st to Nth order C-change values are repeatedly calculated by using the aligned random data and the aligned two-dimensional conjugate sequence for noise addition.

[0059] The specific method for determining whether the candidate input two-dimensional data sequence has conjugacy through C-transformation includes:

[0060] Arbitrarily selecting two groups of data from the candidate input two-dimensional data sequence as the first input data and the second input data respectively;

[0061] Calculating the C-transformation value by using the first input data and the third noise to obtain a third C-transformation value;

[0062] Calculating the C-transformation value by using the second input data and the fourth noise to obtain a fourth C-transformation value;

[0063] Calculating the average of the third C-transformation value and the average of the fourth C-transformation value, and determining whether the average of the third C-transformation value and the average of the fourth C-transformation value are within a set range;

[0064] If yes, the first input data and the second input data are conjugate, and step S5 is performed;

[0065] If no, the first input data and the second input data are not conjugate, and if not conjugate, the step S3 is continuously returned to be executed until a new candidate input two-dimensional data sequence is generated.

[0066] The new candidate input two-dimensional data sequence represents a "natural" entity, and the new candidate input two-dimensional data sequence is continuously exchanged and paired to obtain a paired data sequence, and it is determined whether the paired data sequence converges, if not, the step S3 is continuously returned to be executed, and if yes, the step S6 is performed. The probability of each output short sentence sequence is calculated through step S7, and a certain number of output short sentence sequences with high probability are selected and connected as an output sequence. The output sequence and the embedded numerical sequence are compared and optimized to obtain the parameter value of the artificial intelligence dialogue system with minimum error.

[0067] In the embodiment of the present application, the data generated by using the machine learning algorithm and MCMC (Monte Carlo) sampling has the same information entropy as single data and N data, that is, H(S) = NH(S), so as to ensure that the operation data complies with the exchange principle. By continuously exchanging positions, all possible combinations can be generated.

[0068] In MCMC sampling, random data is used to represent all data. According to the law of large numbers, as long as a large number of random numbers are operated regularly, a convergent result, E(f(A)), will be obtained, where A is a random number and f is an operation rule. Bringing the statistical value of each random data into the same operation will also obtain a convergent result, E(f(A)) = f(E(A)). This is because the relationship H(S) = NH(S) makes the single character and all character sequences have the same information entropy. f(A) can be regarded as a Gödel expression (f, A) similar to the typical sequence. Since the operation rule f has the same expression as the typical sequence, (f, A) will certainly obtain the same statistical law f(E(A)).

[0069] Conjugate refers to a pair of pairing according to certain rules, that is, twin. In mathematics, there are conjugate complex numbers, conjugate roots, conjugate hyperbolas, conjugate matrices, etc. These can be achieved by pairing functions, pairing matrices, pairing matrix transpositions, pairing recursive functions, pairing matrices, and pairing matrix transpositions. Conjugate is also a special relationship between two vectors. Let A be an n x n symmetric positive definite matrix, and vector p, p ∈ R. If it satisfies the condition (p)Ap = 0, then p and p are conjugate directions with respect to A, or p and p are conjugate with respect to A. Generally, for a non-zero vector group p, p, …, p ∈ R, if it satisfies the condition: (p)Ap = 0 (i ≠ j, i, j = 1, 2, …, n), then the vector group is conjugate with respect to A. In the search process, similar to the previous f and p transfer functions, according to the properties of conjugate directions, a one-dimensional search is made along a group of directions conjugate to Q, which can guarantee that the minimum point of the quadratic function is obtained within at most n steps. Conjugate is essentially an extension of symmetry, which is easier to judge with functions than symmetry. One judgment method is to swap the parameters x and y of the function f and see if the equation still holds. That is, whether f(x, y) == f(y, x), or whether f(x, y) == 1 / f(y, x). When a sequence A is randomly mixed with another conjugate sequence B, if the conjugate property is still satisfied, such as f(x A , y B ) == f(y B , x A ) and f(x A , x B ) == f(x B , x A ), then a double conjugate relationship is brought about, that is, sequence A also satisfies the conjugate property with sequence B as the condition, then any two elements x A , x B of sequence A satisfy the conjugate property, which is similar to the idea of extended Gibbs sampling, that is, in the extreme case, there must exist two conjugate functions p(x, y) and g(y, x), which have symmetry p(x, y)g(y *x) == p(x, y * ) g(y, x) and p(x, y) g(y, x * ) == p(x * , y) g(y, x), which is the necessary condition of conjugate under the extreme condition, so the sequence A satisfies the fine balance condition. The C-transform is used to judge whether the alternative input two-dimensional data sequence is conjugate or not, and the paired data sequence is obtained by performing sentence exchange pairing on the alternative input two-dimensional data sequence, and whether the paired data converges is judged, that is, whether the input and output remain unchanged is constantly evaluated, and keeping the conjugate unchanged will increase the adaptability of the construction method of the embodiment.

[0070] Since the process combines two different random and deterministic processes, it reflects the relationship between the random sequence of infinite possibilities and the finite parameter determination and identification calculation, which is equivalent to replacing a large number of Monte Carlo sampling with a finite process, or recursively inducing a finite number of parameters with an infinite coefficient. The typical sequence generated in this way is always symmetric and orthogonal, as long as the fine balance relationship at the segmentation is satisfied (that is, the boundary condition is optimized, or the identified parameters satisfy the symmetric or cancellation relationship, which is equivalent to the optimization of the boundary condition, and each change or left boundary and right boundary corresponds to an unchanged result), the mutual intersection coefficient of each segment is a sharp pulse with a phase shift of zero or itself as the center, and the estimates of each segment are also the same, that is, the conjugate is satisfied.

[0071] As long as the interpolation is fine enough, the list obtained from the typical sequence (with randomness, stationarity and convergence) becomes a large enough sequence list (such as a determinant), which must also reflect the infinite properties of each segment.

[0072] The embodiment of the application is based on the best Monte Carlo Markov chain (MCMC) and Gibbs sampling method, and establishes a segmented and conjugate Markov chain transition process under random maximization or other extreme conditions through optimized character encoding, and then performs a similar "nature", that is, an assumed entity or intermediate variable with certain prior knowledge of the target or the most widely affected entity or intermediate variable. In fact, it is a process of rejecting and accepting conjugate data by adding noise and C-transform, constantly evaluating whether the input and output can remain unchanged when the assumed "natural" entity or intermediate variable changes, keeping the conjugate unchanged will increase the adaptability of the system, simplify the optimization rules, and improve the efficiency of the system, and efficiently obtain the typical distribution parameters of the input and output of the machine learning to be measured.

[0073] The C-transform-based artificial intelligence dialogue system construction method provided by the embodiment of the application can improve the efficiency of Chinese sentence analysis by learning the code value of Chinese characters, automatically segmenting, and establishing the relationship between corresponding phrases.

[0074] In the first embodiment described above, a C-transform-based artificial intelligence dialogue system construction method is provided, and a C-transform-based artificial intelligence dialogue system construction device is also provided. Figure 2 Fig. 2 is a structural schematic diagram of a C-transform-based artificial intelligence dialogue system construction device according to the second embodiment of the application.

[0075] As shown in Fig. 2, a structural schematic diagram of a C-transform-based artificial intelligence dialogue system construction device according to another embodiment of the application is shown. Figure 2 As shown in Fig. 2, a structural schematic diagram of a C-transform-based artificial intelligence dialogue system construction device according to another embodiment of the application is shown.

[0076] The acquisition module is configured to acquire demonstration data of the artificial intelligence dialogue system, map each character and the corresponding code value by using the question part of the demonstration data, and obtain an input position embedding value sequence in combination with the position of each character.

[0077] The calculation module is configured to calculate the input position embedding value sequence by using the Tower of Hanoi algorithm to obtain a two-dimensional conjugate sequence.

[0078] The fusion module is configured to acquire random data, perform segmentation processing on the random data to obtain segmented random data, and fuse the segmented random data and the two-dimensional conjugate sequence by using C-transform to generate a candidate input two-dimensional data sequence.

[0079] The judgment module is configured to judge whether the candidate input two-dimensional data sequence has conjugacy by using C-transform, analyze the candidate input two-dimensional data sequence by using the analysis module if the candidate input two-dimensional data sequence has conjugacy, and perform fusion processing on the candidate input two-dimensional data sequence by using the fusion module if the candidate input two-dimensional data sequence does not have conjugacy.

[0080] The analysis module performs short sentence exchange pairing on the alternative input two-dimensional data sequence to obtain a paired data sequence, judges whether the paired data sequence converges, if converges, performs random segmentation processing by using a segmentation module, and if does not converge, performs fusion processing by using a fusion module;

[0081] The segmentation module performs random segmentation on the reply part in the demonstration data to obtain an embedding numerical sequence of each segment output short sentence;

[0082] The processing module calculates the number of corresponding output short sentence sequences and the corresponding probability of all output short sentence sequences in the output by using the noise-added data used in the C-transformation process, all word codes of words as learning parameters and the paired data sequence, outputs the short sentence with the maximum probability, and sequentially performs deep learning on all previous part output short sentences and input short sentences to generate the output short sentence of the following part, compares the output sequence formed by connecting all the output short sentence sequences with the maximum probability, and compares the output sequence with the embedding numerical sequence to optimize the parameter value of the artificial intelligence conversation system with the minimum error.

[0083] The deterministic result and the random result of the data generated by the calculation module and the fusion module can generate an optimized result.

[0084] The fusion module includes an alignment unit and a fusion unit, the alignment unit is used for aligning the segmented random data and the two-dimensional conjugate sequence to obtain aligned random data and aligned two-dimensional conjugate sequence;

[0085] The fusion unit is used for fusing the aligned random data and the aligned two-dimensional conjugate sequence.

[0086] The embodiment of the application provides a kind of artificial intelligence conversation system construction device based on C-transform, by learning the code value of Chinese character, automatic segmentation, establish the relationship between corresponding phrase, can improve the efficiency of Chinese sentence analysis. By the C-transform reconstruction code information of noise, the relationship between input short sentence and output short sentence is constantly generated new output short sentence by using deep learning, all the maximum output short sentence is linked to obtain output sequence, the embedding numerical sequence obtained by the output sequence and the demonstration sentence segmentation is compared and optimized, and the final conversation output result is obtained. The conversation output result obtained is more accurate, and the sentence analysis efficiency is high.

[0087] As Figure 3As shown, another embodiment of the application also provides a structural diagram of an intelligent terminal, which includes a processor, an input device, an output device and a memory, the processor, the input device, the output device and the memory are connected with each other, the memory is used for storing a computer program, the computer program includes program instructions, the processor is configured to invoke the program instructions and execute the method described in the above embodiment.

[0088] It should be understood that, in the embodiments of the application, the processor can be a central processing unit (CPU), and the processor can 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. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor.

[0089] The input device can include a touchpad, a fingerprint acquisition sensor (used for acquiring fingerprint information and direction information of a fingerprint of a user), a microphone, etc., and the output device can include a display (LCD, etc.), a speaker, etc.

[0090] The memory can include read-only memory and random access memory, and provide instructions and data for the processor. A portion of the memory can also include non-volatile random access memory. For example, the memory can also store device type information.

[0091] In specific implementations, the processor, the input device and the output device described in the embodiments of the application can execute the implementation manners described in the method embodiments provided by the embodiments of the application, and can also execute the implementation manners of the system embodiments described in the embodiments of the application, which will not be described here.

[0092] The application also provides an embodiment of a computer readable storage medium, the computer storage medium stores a computer program, the computer program includes program instructions, the program instructions, when executed by a processor, cause the processor to execute the method described in the above embodiment.

[0093] The computer readable storage medium can be an internal storage unit of the terminal, such as a hard disk or a memory of the terminal. The computer readable storage medium can also be an external storage device of the terminal, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. equipped on the terminal. Further, the computer readable storage medium can include both the internal storage unit and the external storage device of the terminal. The computer readable storage medium is used to store the computer program and other programs and data required by the terminal. The computer readable storage medium can also be used to temporarily store data that has been output or will be output.

[0094] Those skilled in the art can understand that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be realized in electronic hardware, computer software or a combination of both. In order to clearly illustrate the interchangeability of hardware and software, the components and steps of each example have been described in the above description in general terms. Whether the functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. A skilled person can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0095] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the terminal and the units described above can refer to the corresponding processes in the foregoing method embodiments, which will not be described here.

[0096] In several embodiments provided in the present application, it should be understood that the disclosed terminal and method can be implemented in other ways. For example, the above-described device embodiments are only schematic, for example, the division of the units is only a logical function division, and actual implementation can have another division manner, for example, a plurality of units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the units shown or discussed can be indirect coupling or communication connection through some interfaces, devices or units, and can also be electrical, mechanical or other forms of connection.

[0097] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, and are not intended to limit the present application; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that the technical solutions recorded in the foregoing embodiments can still be modified, or some or all of the technical features can be replaced by equivalents; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present application, and they should be covered in the scope of the claims and the description of the present application.

Claims

1. A method for constructing an artificial intelligence dialogue system based on C-transformation, characterized by, The method comprises the following steps: S1: obtaining demonstration data of an artificial intelligence dialogue system, mapping each character and a corresponding character code value by using a question part of the demonstration data, and combining positions of each character to obtain an input position embedding value sequence; S2: calculating the input position embedding value sequence by using a Tamde algorithm to obtain a two-dimensional conjugate sequence; S3: obtaining random data, performing segmentation processing on the random data to obtain segmented random input data, and fusing the segmented random input data and the two-dimensional conjugate sequence by using C-transformation to generate a candidate input two-dimensional data sequence; S4: determining whether the candidate input two-dimensional data sequence has conjugacy by using C-transformation, if yes, performing step S5, and if no, returning to step S3; S5: performing short sentence exchange pairing on the candidate input two-dimensional data sequence to obtain a paired data sequence, determining whether the paired data sequence converges, if no, returning to step S3, and if yes, performing step S6; S6: performing random segmentation on a reply part in the demonstration data to obtain an embedding value sequence of each segment of an output short sentence; S7: calculating a number of corresponding output short sentence sequences and probabilities of all output short sentence sequences in output by using noise-added data used in C-transformation, all character codes of characters as learning parameters, and the paired data sequence, outputting a short sentence with the maximum probability, performing deep learning on all previous part output short sentences and input short sentences in sequence to generate a subsequent part output short sentence, comparing output sequences formed by connecting all output short sentence sequences with the maximum probability together, and comparing the output sequences with the embedding value sequence to obtain parameter values of the artificial intelligence dialogue system with minimized errors; The C-transformation comprises input data and noise, and a specific method of the C-transformation comprises the following steps: assigning the input data to a data variable; setting noise-added data, setting an initial value of a noise amplitude, and setting a step variable to zero; multiplying the noise-added data by the initial value of the amplitude and adding the result to the data variable; performing regularization on the data variable; determining whether the data variable after regularization is white noise, if yes, returning the step variable as output, if not, automatically increasing the step variable by 1, adjusting the noise amplitude, and returning to perform the multiplication of the noise-added data by the initial value of the amplitude and the addition of the result to the data variable to obtain a number of steps required to generate white noise.

2. The method of claim 1, wherein, A specific method of fusing the segmented random input data and the two-dimensional conjugate sequence by using C-transformation comprises the following steps: aligning the segmented random input data and the two-dimensional conjugate sequence to obtain aligned random data and aligned two-dimensional conjugate sequence; fusing the aligned random data and the aligned two-dimensional conjugate sequence.

3. The method of claim 2, wherein, A specific method of aligning the segmented random input data and the two-dimensional conjugate sequence comprises the following steps: calculating 1 to N order C-transformation values by using the segmented random input data and first noise to obtain aligned random data, N being an alignment length; calculating 1 to N order C-transformation values by using the two-dimensional conjugate sequence and second noise to obtain aligned two-dimensional conjugate sequence.

4. The method of claim 3, wherein, A specific method of fusing the aligned random data and the aligned two-dimensional conjugate sequence comprises the following steps: Calculating the first C-transform value by using the aligned random data and the aligned two-dimensional conjugate sequence to do the noise calculation; Calculating the second C-transform value by using the aligned two-dimensional conjugate sequence and the aligned random data to do the noise calculation; Repeating the alternating calculation of each order C-transform value by using the aligned random data and the aligned two-dimensional conjugate sequence to do the noise calculation.

5. The method of claim 4, wherein, The specific method for judging whether the alternative input two-dimensional data sequence has conjugacy through C-transform includes: Arbitrarily selecting two groups of data from the alternative input two-dimensional data sequence as the first input data and the second input data; Calculating the third C-transform value by using the first input data and the third noise to calculate the C-transform value; Calculating the fourth C-transform value by using the second input data and the fourth noise to calculate the C-transform value; Calculating the average of the third C-transform value and the average of the fourth C-transform value, and judging whether the average of the third C-transform value and the average of the fourth C-transform value are within the set range; If yes, the first input data and the second input data are conjugate; If no, the first input data and the second input data are not conjugate. 6.A C-transform-based artificial intelligence conversation system construction apparatus, characterized by comprising: It includes: The acquisition module, the calculation module, the fusion module, the judgment module, the analysis module, the segmentation module and the processing module, wherein, The acquisition module is used to acquire the demonstration data of the artificial intelligence dialogue system, and the mapping of each character and the corresponding character code value is performed by using the question part of the demonstration data, and the input position embedding value sequence is obtained by combining the positions of each character; The calculation module calculates the input position embedding value sequence through the Tomde algorithm to obtain a two-dimensional conjugate sequence; The fusion module acquires random data, and performs segmentation processing on the random data to obtain segmented random input data, and fuses the segmented random input data and the two-dimensional conjugate sequence through C-transform to generate an alternative input two-dimensional data sequence; The judgment module judges whether the alternative input two-dimensional data sequence has conjugacy through C-transform, and if it has conjugacy, the analysis module is used for analysis, and if it does not have conjugacy, the fusion module is used for fusion processing; The analysis module performs short sentence exchange pairing on the alternative input two-dimensional data sequence to obtain a paired data sequence, judges whether the paired data sequence converges, if it converges, the segmentation module is used for random segmentation processing, and if it does not converge, the fusion module is used for fusion processing; The segmentation module randomly segments the reply part in the demonstration data to obtain an embedding value sequence of each segment output short sentence; The processing module calculates the number of corresponding output short sentence sequences and the corresponding probability of all output short sentence sequences in the output by using the noise data used in the C-transform process, the character codes of all characters as learning parameters and the paired data sequence, outputs the short sentence with the maximum probability, and sequentially performs deep learning on all previous part output short sentences and input short sentences to generate the output short sentences of the subsequent part. The output sequence formed by connecting all the output short sentence sequences with the maximum probability is compared with the embedding value sequence to obtain the parameter value of the artificial intelligence dialogue system with the minimum error after optimization; The C-transform includes input data and noise, and the specific method of the C-transform includes: Assign input data to data variable; Set the noise-added data, set the initial value of the noise amplitude, and set the step variable to zero; Multiply the noise-added data by the initial value of the amplitude and add it to the data variable; Regularize the data variable; Determine whether the regularized data variable is white noise, if yes, return the step variable as output, if not, automatically increase the step variable by 1, adjust the noise amplitude, return to execute the step of multiplying the noise-added data by the initial value of the amplitude and adding it to the data variable, and obtain the number of steps required to generate white noise.

7. The apparatus of claim 6, wherein, The fusion module comprises an alignment unit and a fusion unit, the alignment unit is used for aligning the segmented random input data and the two-dimensional conjugate sequence to obtain aligned random data and aligned two-dimensional conjugate sequence; The fusion unit is used for fusing the aligned random data and the aligned two-dimensional conjugate sequence.

8. An intelligent terminal comprising a processor, an input device, an output device and a memory, the processor, the input device, the output device and the memory being connected to each other, the memory being used to store a computer program, the computer program comprising program instructions, characterized in that, The processor is configured to invoke the program instructions to execute the method of any one of claims 1-5.

9. A computer-readable storage medium, characterized in that, The computer readable storage medium stores a computer program, the computer program comprises program instructions, the program instructions make the processor execute the method of any one of claims 1-5 when the processor executes.

Images