Multi-source data latent semantic cognition result prediction method and device

Abstract

The invention discloses a multi-source data latent semantic cognition result prediction method and device, and the method comprises the steps: obtaining multi-source data of a target task field, and extracting the text description of the multi-source data and the corresponding semantic feature information; inputting the text description and the semantic feature information into a latent semantic model to extract latent semantics by using a matrix singular value decomposition algorithm; and processing the latent semantic based on a neural network model to obtain a corresponding explanation. According to dominant features and implicit interaction information of the multi-source data of the target task and typical tasks, latent semantic feature information conforming to the multi-source data of the target task and semantics represented by a hidden layer are effectively mined, and interpretable understanding is achieved.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a method for predicting latent semantic cognitive results of multi-source data. Background Art

[0002] Interpretability refers to the degree to which a model can provide clear and easy-to-understand explanations when making decisions. Research on model interpretability has become a hot topic at scientific research conferences in recent years, because people are not only satisfied with the effects of the model, but also think more about the reasons for the model's effects. Such thinking helps optimize the model and features, and can help better understand the model itself and improve the quality of model services. Deep learning models are uninterpretable due to the number of parameters and the complex methods of extracting and combining features. At present, feature importance analysis is a commonly used model interpretation technique that is used to evaluate the degree of influence of each feature in the model on the prediction results. This can help us understand how the model uses the input data to make decisions. Although feature importance analysis can quickly screen out features that have a greater impact on model performance, reduce computational costs, and is simple to understand and easy to explain. However, it cannot consider the interaction between features. In particular, some information may be lost when fewer features are retained. Summary of the invention

[0003] The main purpose of the present invention is to provide a method and device for predicting latent semantic cognition results of multi-source data to solve the technical problems existing in the related art.

[0004] In order to achieve the above-mentioned purpose, according to the first aspect of the present invention, a method for predicting latent semantic cognitive results of multi-source data is provided, comprising acquiring multi-source data in the target task field, and extracting text descriptions of the multi-source data and corresponding semantic feature information; inputting the text descriptions and semantic feature information into a latent semantic model to extract latent semantics using a matrix singular value decomposition algorithm; and processing the latent semantics based on a neural network model to obtain corresponding explanations.

[0005] Optionally, when extracting the latent semantics, the method includes: constructing a semantic feature information and text description matrix based on the multi-source data text description and the corresponding semantic feature information:

[0006]

[0007] Truncated singular value decomposition of the matrix:

[0008]

[0009] Where k≤n≤m, Uk is an m*k matrix whose columns consist of the first k mutually orthogonal left singular vectors of U, ∑k is a k-order diagonal matrix whose diagonal elements consist of the first k largest singular values ​​of Σ, and Vk is an n*k matrix whose columns consist of the first k mutually orthogonal right singular vectors of V; Uk is used as the semantic vector space; determine the latent semantics:

[0010]

[0011] Optionally, the neural network model includes 1 input layer, 3 hidden layers and 1 output layer, and the number of neurons in each layer is half the number of neurons in the previous layer.

[0012] Optionally, a corresponding latent semantic feature information subset is determined based on the latent semantics, and the latent semantics is interpreted using a similarity calculation function and a deep neural network model algorithm.

[0013] According to a second aspect of the present invention, there is provided a device for predicting cognitive results of latent semantics of multi-source data, comprising a data preprocessing unit for acquiring multi-source data in a target task field, and extracting text descriptions of the multi-source data and corresponding semantic feature information; a latent semantics extraction unit, configured to input the text descriptions and semantic feature information into a latent semantics model to extract latent semantics using a matrix singular value decomposition algorithm; and an interpretation unit, for processing the latent semantics based on a neural network model to obtain corresponding interpretations.

[0014] Optionally, when extracting the latent semantics: constructing a semantic feature information and text description matrix based on the multi-source data text description and the corresponding semantic feature information:

[0015]

[0016] Truncated singular value decomposition of the matrix:

[0017]

[0018] Where k≤n≤m, Uk is an m*k matrix whose columns consist of the first k mutually orthogonal left singular vectors of U, ∑k is a k-order diagonal square matrix whose diagonal elements consist of the first k largest singular values ​​of Σ, and Vk is an n*k matrix whose columns consist of the first k mutually orthogonal right singular vectors of V; Uk is used as the semantic vector space;

[0019] Determine the implicit semantics:

[0020]

[0021] Optionally, the neural network model includes 1 input layer, 3 hidden layers and 1 output layer, and the number of neurons in each layer is half the number of neurons in the previous layer.

[0022] Optionally, a corresponding latent semantic feature information subset is determined based on the latent semantics, and the latent semantics is interpreted using a similarity calculation function and a deep neural network model algorithm.

[0023] According to a third aspect of the present invention, there is provided a computer-readable storage medium storing computer instructions, wherein the computer instructions are used to cause the computer to execute the method described in any one of the implementations of the first aspect.

[0024] According to a fourth aspect of the present invention, there is provided an electronic device comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor executes the method described in any one implementation of the first aspect.

[0025] The present embodiment provides a method and device for predicting the cognitive results of latent semantics of multi-source data, wherein the method includes obtaining multi-source data in the field of a target task, extracting text descriptions of the multi-source data and corresponding semantic feature information; inputting the text descriptions and semantic feature information into a latent semantic model to extract latent semantics using a matrix singular value decomposition algorithm; and processing the latent semantics based on a neural network model to obtain corresponding explanations. Based on the explicit features and implicit interaction information between the multi-source data of the target task and typical tasks, the latent semantic feature information that meets the multi-source data of the target task and the semantics represented by the hidden layer are effectively mined, achieving interpretable understanding. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0027] Figure 1 is a flow chart of a method for predicting latent semantic cognition results of multi-source data according to an embodiment of the present invention;

[0028] Figure 2 is a schematic diagram of an electronic device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0029] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0030] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so as to describe the embodiments of the present invention described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0031] It should be noted that, in the absence of conflict, the embodiments of the present invention and the features in the embodiments can be combined with each other. The present invention will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0032] According to an embodiment of the present invention, a method for predicting latent semantic cognition results of multi-source data is provided. Figure 1 As shown, it includes the following steps 101 to 103:

[0033] Step 101: Acquire multi-source data in the target task domain, and extract text descriptions of the multi-source data and corresponding semantic feature information.

[0034] In this step, multi-source data in the target task domain is obtained. Based on the domain-related knowledge and according to the target task characteristics and attributes, the text description of the multi-source data of the target task and the semantic feature information contained therein are extracted to obtain the text description of the multi-source data in the target task domain and the corresponding semantic feature information set. The multi-source target task domain data set includes statistical figures, images, text, speech and other data, as well as target recognition task data, target task prediction data and text parsing task data in complex analysis scenarios. Here, it is necessary to take the multi-source target task data as the core, and conduct semantic feature information mining and cognitive combing on the subject, target task type, business attributes and spatiotemporal attributes of the above data.

[0035] Step 102: Input the text description and semantic feature information into a latent semantic model to extract latent semantics using a matrix singular value decomposition algorithm.

[0036] In this embodiment, based on the above multi-source data text description and the corresponding semantic feature information set, a latent semantic model is adopted, and a matrix singular value decomposition algorithm is used to extract the semantic relationship between the multi-source data text description based on the target task and the corresponding semantic feature information in the above set, and obtain the potential semantic representation information of its hidden layer. Singular value decomposition (SVD) is a matrix factorization method, which is to represent a complex matrix in the form of multiplying smaller and simpler sub-matrices, and is mainly used to find the "pattern" hidden in a large amount of data.

[0037] Step 103: Process the latent semantics based on the neural network model to obtain corresponding explanations.

[0038] This embodiment starts from the perspective of human cognitive semantics of the target task, and establishes a multi-source data latent semantic interpretable computing model in the target and task domain by mapping the target task neurons and semantic relationships. That is, the neurons inside the model are semantically interpreted, and the semantic feature analysis and cognitive computing tasks of the target task are completed, thus solving the problem that deep neural networks are uninterpretable.

[0039] As an optional implementation of this embodiment, when extracting the latent semantics, the method includes: constructing a semantic feature information and text description matrix based on the multi-source data text description and the corresponding semantic feature information:

[0040] Truncated singular value decomposition of the matrix:

[0041]

[0042] Where k≤n≤m, Uk is an m*k matrix whose columns consist of the first k mutually orthogonal left singular vectors of U, ∑k is a k-order diagonal matrix whose diagonal elements consist of the first k largest singular values ​​of Σ, and Vk is an n*k matrix whose columns consist of the first k mutually orthogonal right singular vectors of V; Uk is used as the semantic vector space; determine the latent semantics:

[0043]

[0044] In this optional implementation, when determining latent semantics, (1) a semantic feature information (generally words)-text description matrix of multi-source data in the target task domain is constructed. For a given text description set D and a semantic feature information (word) set W, a semantic feature information (word)-text description matrix X can be obtained:

[0045]

[0046] (2) Truncated singular value decomposition

[0047]

[0048] Because it is a truncated singular value decomposition, k≤n≤m, Uk is an m*k matrix whose columns consist of the first k mutually orthogonal left singular vectors of U, ∑k is a k-order diagonal matrix whose diagonal elements consist of the first k largest singular values ​​of Σ, and Vk is an n*k matrix whose columns consist of the first k mutually orthogonal right singular vectors of V.

[0049] (3) The above semantic vector space is obtained. Here, the semantic vector space is Uk.

[0050] (4) Establish a semantic feature information (word)-text description matrix. Here, the semantic feature information (word)-text description matrix is ​​the product of Σk and Vk:

[0051]

[0052] According to the latent semantic representation information vector of its hidden layer, the approximate nearest neighbor search technology is used to recall the latent semantic feature information in the above vector library, and establish a latent semantic feature information subset.

[0053] As an optional implementation of this embodiment, the neural network model includes 1 input layer, 3 hidden layers and 1 output layer, and the number of neurons in each layer is half the number of neurons in the previous layer.

[0054] In this optional implementation, based on the above-mentioned latent semantic feature information subset, a multi-source data latent semantic interpretable cognitive model in the target task domain is established using a similarity calculation function and a deep neural network model algorithm. That is, the neurons in the hidden layer that truly represent the semantic feature information are decomposed to obtain the encoding of the decoupled neurons and the corresponding semantic feature information. The similarity calculation function is used to calculate the degree of fit between the hidden layer features of a certain hidden layer under a single data and the multi-source target task semantic feature information interpretation data set, and the semantic feature information with a degree of fit greater than the set threshold is selected as the alternative semantics of the hidden layer features under a single data. Finally, the semantic feature information that appears most frequently in the alternative semantic feature information set corresponding to the hidden layer features under all data is counted and defined as the semantic feature information represented by the hidden layer, and the representation of the potential semantic investment information of a certain hidden layer of the deep learning model to be interpreted is obtained, and the latent semantic interpretation of the deep learning model is completed.

[0055] It should be noted that semantic feature analysis can better reveal the reasons why a syntactic combination is established. For example, "willow, poplar, jujube" is considered a meaningful combination because they share the category feature of [+ tree], while "willow, sun, waterfall" is meaningless because they lack common category features; secondly, semantic feature analysis can reveal the more hidden semantic features of certain words. For example, the verb "stay" itself does not contain the meaning of "give". But when it comes to "reserving a seat", it is possible to obtain the meaning of "give".

[0056] For the data to be predicted, the semantic label information (semantic feature information) of the multi-source data of the typical task to be explained is extracted, and the above model is used to perform semantic interpretation of the internal neurons of the deep learning model of the multi-source typical task, that is, the latent semantic interpretation of the model, to complete the understanding of the "black box" model of the multi-source typical task data.

[0057] The method for constructing a data latent semantic interpretable cognitive model based on semantic feature analysis in this embodiment can effectively mine the latent semantic feature information that conforms to the multi-source data of the target task and the semantics represented by the hidden layer according to the explicit features and implicit interaction information of the multi-source data of the target task and the typical task, thereby realizing interpretable understanding. By semantically interpreting the neurons inside the deep learning model of the multi-source typical task, that is, the latent semantic interpretation of the model, the understanding of the "black box" model of the multi-source typical task data is completed, the credibility of the model results is improved, and the intuitive cognition of people to the interpretable results is enhanced. The latent semantic model is used to mine the latent semantic features of the multi-source data of the target task. The model can automatically cluster the target task and the typical task items based on the latent semantic features of the multi-source data of the target task, and model the multi-source data of the target task as a model based on the latent semantic feature information vector of the target task, which can more accurately calculate the similarity of the target task.

[0058] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0059] According to an embodiment of the present invention, a multi-source data latent semantic cognitive result prediction device is also provided, including a data preprocessing unit, used to obtain multi-source data in the target task field, and extract the multi-source data text description and corresponding semantic feature information; a latent semantic extraction unit, configured to input the text description and semantic feature information into a latent semantic model to extract latent semantics using a matrix singular value decomposition algorithm; an interpretation unit, used to process the latent semantics based on a neural network model to obtain corresponding interpretations.

[0060] As an optional implementation of this embodiment, when extracting the latent semantics: constructing a semantic feature information and text description matrix based on the multi-source data text description and the corresponding semantic feature information:

[0061]

[0062] Truncated singular value decomposition of the matrix:

[0063]

[0064] Where k≤n≤m, Uk is an m*k matrix whose columns consist of the first k mutually orthogonal left singular vectors of U, ∑k is a k-order diagonal matrix whose diagonal elements consist of the first k largest singular values ​​of Σ, and Vk is an n*k matrix whose columns consist of the first k mutually orthogonal right singular vectors of V; Uk is used as the semantic vector space; determine the latent semantics:

[0065]

[0066] As an optional implementation of this embodiment, the neural network model includes 1 input layer, 3 hidden layers and 1 output layer, and the number of neurons in each layer is half the number of neurons in the previous layer.

[0067] As an optional implementation method of the present embodiment, a corresponding latent semantic feature information subset is determined based on the latent semantics, and the latent semantics is interpreted using a similarity calculation function and a deep neural network model algorithm.

[0068] According to an embodiment of the present invention, the present invention also provides an electronic device, which includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can implement the method described in any of the above embodiments when executing.

[0069] According to an embodiment of the present invention, the present invention further provides a readable storage medium, which stores computer instructions, and the computer instructions are used to enable a computer to implement the method described in any of the above embodiments when executed.

[0070] According to an embodiment of the present invention, the present invention further provides a computer program product, which can implement the method described in any of the above embodiments when executed by a processor.

[0071] Figure 2A schematic block diagram of an example electronic device 300 that can be used to implement an embodiment of the present invention is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices, and other similar computing devices.

[0072] like Figure 2 As shown, the electronic device 300 includes a computing unit 301, which can perform various appropriate actions and processes according to a computer program stored in a read-only memory (ROM) 302 or a computer program loaded from a storage unit 308 into a random access memory (RAM) 303. In the RAM 303, various programs and data required for the operation of the electronic device 300 can also be stored. The computing unit 301, the ROM 302, and the RAM 303 are connected to each other via a bus 304. An input / output (I / O) interface 305 is also connected to the bus 304.

[0073] A number of components in the electronic device 300 are connected to the I / O interface 305, including: an input unit 306, such as a keyboard, a mouse, etc.; an output unit 307, such as various types of displays, speakers, etc.; a storage unit 308, such as a disk, an optical disk, etc.; and a communication unit 309, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 309 allows the electronic device 300 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0074] The computing unit 301 may be a variety of general and / or special processing components with processing and computing capabilities. Some examples of the computing unit 301 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, digital signal processors (DSPs), and any appropriate processors, controllers, microcontrollers, etc. The computing unit 301 performs the various methods and processes described above, such as the object matching method. For example, in some embodiments, the object matching method may be implemented as a computer software program, which is tangibly included in a machine-readable medium, such as a storage unit 308. In some embodiments, part or all of the computer program may be loaded and / or installed on the electronic device 300 via the ROM 302 and / or the communication unit 309. When the computer program is loaded into the RAM 303 and executed by the computing unit 301, one or more steps of the method described above may be performed.

[0075] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), load programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include: being implemented in one or more computer programs, which may be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general programmable processor, which may receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0076] The program code for implementing the method of the present invention can be written in any combination of one or more programming languages. These program codes can be provided to a processor or controller of a general-purpose computer, a special-purpose computer or other programmable data processing device, so that the program code, when executed by the processor or controller, enables the functions / operations specified in the flow chart and / or block diagram to be implemented. The program code can be executed entirely on the machine, partially on the machine, partially on the machine as a stand-alone software package and partially on a remote machine, or entirely on a remote machine or server.

[0077] In the context of the present invention, a machine-readable medium may be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, device, or equipment. A machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, or equipment, or any suitable combination of the foregoing. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

Claims

1. A method for predicting latent semantic cognition results of multi-source data, characterized in that: include: Obtain multi-source data in the target task domain and extract text descriptions of the multi-source data and corresponding semantic feature information; Inputting the text description and semantic feature information into a latent semantic model to extract latent semantics using a matrix singular value decomposition algorithm; The latent semantics are processed based on a neural network model to obtain corresponding explanations.

2. The method for predicting latent semantic cognition results of multi-source data according to claim 1 is characterized in that: When extracting the latent semantics, the method includes: Constructing a semantic feature information and text description matrix based on the multi-source data text description and the corresponding semantic feature information: Truncated singular value decomposition of the matrix: Where k≤n≤m, Uk is an m*k matrix whose columns consist of the first k mutually orthogonal left singular vectors of U, ∑k is a k-order diagonal square matrix whose diagonal elements consist of the first k largest singular values ​​of Σ, and Vk is an n*k matrix whose columns consist of the first k mutually orthogonal right singular vectors of V; Uk is used as the semantic vector space; Determine the implicit semantics:

3. The method for predicting latent semantic cognition results of multi-source data according to claim 2 is characterized in that: The neural network model includes 1 input layer, 3 hidden layers and 1 output layer, and the number of neurons in each layer is half of the number of neurons in the previous layer.

4. The method for predicting latent semantic cognition results of multi-source data according to claim 3 is characterized in that: Based on the latent semantics, a corresponding latent semantic feature information subset is determined, and the latent semantics is interpreted using a similarity calculation function and a deep neural network model algorithm.

5. A multi-source data latent semantics cognitive result prediction device, characterized in that: include: A data preprocessing unit is used to obtain multi-source data in the target task field and extract text descriptions of the multi-source data and corresponding semantic feature information; A latent semantics extraction unit is configured to input the text description and semantic feature information into a latent semantics model to extract latent semantics using a matrix singular value decomposition algorithm; The interpretation unit is used to process the latent semantics based on a neural network model to obtain corresponding interpretations.

6. The multi-source data latent semantics cognitive result prediction device according to claim 5, characterized in that: When extracting the implicit semantics: Constructing a semantic feature information and text description matrix based on the multi-source data text description and the corresponding semantic feature information: Truncated singular value decomposition of the matrix: Where k≤n≤m, Uk is an m*k matrix whose columns consist of the first k mutually orthogonal left singular vectors of U, ∑k is a k-order diagonal square matrix whose diagonal elements consist of the first k largest singular values ​​of Σ, and Vk is an n*k matrix whose columns consist of the first k mutually orthogonal right singular vectors of V; Uk is used as the semantic vector space; Determine the implicit semantics:

7. The multi-source data latent semantics recognition result prediction device according to claim 6, characterized in that: The neural network model includes 1 input layer, 3 hidden layers and 1 output layer, and the number of neurons in each layer is half of the number of neurons in the previous layer.

8. The multi-source data latent semantics recognition result prediction device according to claim 3, characterized in that: Based on the latent semantics, a corresponding latent semantic feature information subset is determined, and the latent semantics is interpreted using a similarity calculation function and a deep neural network model algorithm.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable the computer to execute the method according to any one of claims 1 to 4.

10. An electronic device, characterized in that: include: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor executes the method described in any one of claims 1 to 4.

Images