A low-coupling sentiment analysis algorithm library based on the mindspire framework and an intelligent terminal

Abstract

This invention discloses a loosely coupled sentiment analysis algorithm library based on the Ascend framework. The library includes: algorithm modules, exposed interfaces coupled to the algorithm modules, and a unified calling interface coupled to the exposed interfaces. The algorithm modules are used to build and train the algorithms in the library. Each algorithm module includes a data module, a model module, and a training module. The exposed interfaces provide a calling interface for each algorithm to the unified calling interface. The unified calling interface includes hyperparameter configuration information and calling logic. This invention builds the algorithm library based on the domestic Ascend framework. By preserving the integrity of each algorithm, it reduces the coupling between different algorithms, allowing each algorithm to be self-contained while achieving unified scheduling through an exposed interface. This results in a cross-platform, unified architecture, unified calling interface, and easily extensible algorithm library.

Description

Technical Field

[0001] This invention relates to the field of computer technology, specifically to a loosely coupled sentiment analysis algorithm library and smart terminal based on the Ascend framework. Background Technology

[0002] In today's society, artificial intelligence (AI) technology has deeply integrated into all aspects of people's daily lives and is closely related to everyone. However, the hardware and software foundations and algorithms of this technology, which is applied on such a large scale to society, are mostly from foreign manufacturers. Specifically, the hardware mainly uses GPUs manufactured by NVIDIA, and the software frameworks are mostly PyTorch (a domestically developed deep learning framework) or TensorFlow (a deep learning framework developed by Google). Although the emergence of domestically produced AI chips and deep learning frameworks has led to rapid development in both hardware platforms and software technology for deep learning algorithms, there is still a lack of algorithm libraries based on domestically developed deep learning frameworks in the construction of the AI ​​ecosystem.

[0003] Existing sentiment analysis algorithm libraries often tend towards moderate or high coupling, no longer preserving the independence of the algorithms. In moderately coupled libraries, the basic operators of each algorithm are typically extracted to form a separate functional module. Ideally, if the library needs to be expanded, only the basic operators of the functional module need to be used to build the algorithm. However, when the new algorithm differs significantly from existing algorithms, in order to integrate it into the library, developers need to extract the basic operators of the algorithm and integrate them into the functional module themselves. This operation requires developers to understand all the current functional modules and the subtle differences between existing and new functional modules, making the expansion and maintenance of the algorithm library too difficult. In highly coupled algorithm libraries, the entire library is abstracted into several different modules based on the algorithm's input, output, and model. These modules generally cover the complete algorithm execution process. The advantage of this modular design is that the algorithm's execution process is divided and unified, reducing the complexity of algorithm implementation. However, this type of library requires developers to first learn the design of each module and have a full understanding of the entire library before they can use its provided interfaces to build new algorithms, resulting in a high learning cost. In addition, this type of library is not very flexible; if special training methods are involved, the interface may not be applicable.

[0004] Therefore, existing technologies still need to be improved and developed. Summary of the Invention

[0005] The technical problem to be solved by this invention is to provide a low-coupling sentiment analysis algorithm library and smart terminal based on the Ascend framework, which addresses the above-mentioned deficiencies of the prior art. This invention aims to solve the problems in the construction of the artificial intelligence ecosystem, such as the lack of algorithm libraries based on domestic deep learning frameworks, the high difficulty in expanding and maintaining medium-coupling algorithm libraries, the high learning cost, low flexibility, and unsuitable interfaces of highly coupled algorithm libraries.

[0006] The technical solution adopted by this invention to solve the technical problem is as follows:

[0007] In a first aspect, the present invention provides a loosely coupled sentiment analysis algorithm library based on the Ascend framework, wherein the loosely coupled sentiment analysis algorithm library includes: an algorithm module, an exposed interface coupled to the algorithm module, and a unified calling interface coupled to the exposed interface; wherein the algorithm module is used to construct and train the algorithms in the loosely coupled sentiment analysis algorithm library; the algorithm module includes a data module, a model module, and a training module; the exposed interface is used to provide a calling interface for each algorithm to the unified calling interface; the unified calling interface includes hyperparameter configuration information and calling logic.

[0008] In one implementation, the data module includes a data download submodule, a data processing submodule, and a dataset construction submodule; the data module is used to process the data input from the algorithm library and output the processed data to the corresponding module.

[0009] In one implementation, the data download submodule is used to download data, the data processing submodule is used to process all data content related to the algorithm, and the dataset construction submodule is used to construct the dataset interface.

[0010] In one implementation, the model module includes: a model architecture submodule, a model construction submodule, and a model parameter configuration submodule; the model module is used for model architecture design, algorithm code construction, and parameter configuration.

[0011] In one implementation, the model architecture submodule is used for model architecture design, the model construction submodule is used for algorithm code construction, and the model parameter configuration submodule is used for configuring model parameters.

[0012] In one implementation, the training module includes a model training submodule, a model evaluation submodule, and a model inference submodule; the training module is used for model training, evaluation, and inference.

[0013] In one implementation, the model training submodule is used to train the model, the model evaluation submodule is used to evaluate the training effect of the model, and the model inference submodule is used to optimize the inference of the model.

[0014] In one implementation, the unified invocation interface is used to obtain invocation parameters, and based on the invocation parameters and the invocation logic, to uniformly invoke any algorithm through the exposed interface of each algorithm in the loosely coupled sentiment analysis algorithm library.

[0015] In one implementation, the unified calling interface is used to search for the hyperparameters of any algorithm in the loosely coupled sentiment analysis algorithm library based on the hyperparameter configuration information.

[0016] In one implementation, the loosely coupled sentiment analysis algorithm library is built on the Ascend framework.

[0017] In one implementation, the loosely coupled sentiment analysis algorithm library supports multiple computing platforms, including NPU, GPU, and CPU.

[0018] Secondly, embodiments of the present invention also provide a smart terminal, wherein the smart terminal includes a memory, a processor, and the Ascend framework-based low-coupling sentiment analysis algorithm library stored in the memory and capable of running on the processor, the processor running the Ascend framework-based low-coupling sentiment analysis algorithm library as described in any of the above embodiments.

[0019] Beneficial Effects: Compared with existing technologies, this invention provides a loosely coupled sentiment analysis algorithm library based on the Ascend framework. The library includes: algorithm modules, exposed interfaces coupled to the algorithm modules, and a unified calling interface coupled to the exposed interfaces. The unified calling interface improves the library's readability by constructing the library based on a unified architecture. Specifically, the unified calling interface is used to obtain calling parameters and, based on these parameters and the calling logic, to uniformly call any algorithm through the exposed interface of each algorithm in the library. This reduces coupling between different algorithms by preserving the integrity of each algorithm, allowing each algorithm to function independently while achieving unified scheduling through a single exposed interface. This ensures low coupling within the library and reduces the difficulty of expansion, maintenance, and learning. The loosely coupled sentiment analysis algorithm library is built on the Ascend framework, realizing an algorithm library based on a domestic deep learning framework. The library supports multiple computing platforms, including NPU, GPU, and CPU, thus achieving cross-platform compatibility, a unified architecture, a unified calling interface, and easy extensibility. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 This is a schematic diagram of a loosely coupled sentiment analysis algorithm library based on the Ascend framework provided in an embodiment of the present invention.

[0022] Figure 2 This is a schematic diagram of the algorithm module provided in an embodiment of the present invention.

[0023] Figure 3 This is a block diagram illustrating the internal structure of a smart terminal provided in an embodiment of the present invention. Detailed Implementation

[0024] To make the objectives, technical solutions, and effects of this invention clearer and more explicit, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0025] Those skilled in the art will understand that, unless specifically stated otherwise, the singular forms “a,” “an,” “the,” and “the” used herein may also include the plural forms. It should be further understood that the term “comprising” as used in this specification means the presence of the stated features, integers, steps, operations, elements, and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof. It should be understood that when we say an element is “connected” or “coupled” to another element, it can be directly connected or coupled to the other element, or there may be intermediate elements. Furthermore, “connected” or “coupled” as used herein can include wireless connections or wireless coupling. The term “and / or” as used herein includes all or any units and all combinations of one or more associated listed items.

[0026] It will be understood by those skilled in the art that, unless otherwise defined, all terms used herein (including technical and scientific terms) have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. It should also be understood that terms such as those defined in general dictionaries should be understood to have the same meaning as in the context of the prior art, and should not be interpreted in an idealized or overly formal sense unless specifically defined as herein.

[0027] To address the shortcomings of the domestic AI ecosystem, we propose a loosely coupled sentiment analysis algorithm library based on the Ascend framework. This library supports cross-platform compatibility, a unified architecture, a unified API, and easy extensibility. Thanks to the cross-platform capabilities of the domestic deep learning framework Ascend, our library can run simultaneously on NVIDIA GPUs and Huawei Ascend NPUs, achieving complete domestic support from hardware to framework. Through research on the architectures of numerous sentiment analysis algorithms, we abstract a unified architecture for sentiment analysis algorithms and build the library based on this unified architecture to improve readability. Furthermore, to achieve unified scheduling of algorithms within the library, we designed a loosely coupled unified API. This ensures the independence of each algorithm while enabling unified management, improving usability. This loosely coupled unified API also brings the advantage of easy extensibility; as long as an algorithm can run independently, it can be integrated into the library by exposing its interface.

[0028] Exemplary methods

[0029] This embodiment provides a loosely coupled sentiment analysis algorithm library based on the Ascend framework. For example... Figure 1 As shown, the loosely coupled sentiment analysis algorithm library includes: an algorithm module 10, an exposed interface 20 coupled to the algorithm module, and a unified calling interface 30 coupled to the exposed interface; wherein, the algorithm module 10 is used to construct and train the algorithms in the loosely coupled sentiment analysis algorithm library; as shown Figure 2 As shown, the algorithm module 10 includes a data module, a model module, and a training module; the exposed interface 20 is used to provide a calling interface for each algorithm to the unified calling interface; the unified calling interface 30 includes hyperparameter configuration information and calling logic.

[0030] In natural language processing, sentiment analysis generally refers to determining the emotional state expressed by a piece of text. This text can be a sentence, a paragraph, or a document. Emotional states can be two-fold (positive, negative), such as (happy, sad); or three-fold (positive, negative, neutral), etc. Sentiment analysis algorithms have a wide range of applications, such as categorizing user reviews on shopping websites (Amazon, Tmall, Taobao, etc.), travel websites, and movie review websites into positive and negative reviews; or analyzing user experiences with a product by capturing user reviews and performing sentiment analysis.

[0031] Specifically, the loosely coupled sentiment analysis algorithm library in this embodiment adopts a unified architecture, that is, each algorithm is decomposed into a unified format based on the typical structure of sentiment analysis algorithms to enhance the readability of the algorithm library. The unified architecture design mainly includes two parts: the internal structure of the algorithm and the interaction between the algorithm and the unified calling interface. That is, the algorithm module 10 is used to integrate all algorithms in the loosely coupled sentiment analysis algorithm library. To preserve the integrity of the algorithms, each algorithm is self-contained and does not need to be split. The algorithm module 10 interacts with the unified calling interface 30 through the exposed interface 20. The advantage of this design is that it preserves the integrity of each algorithm, but at the same time it can interact with the unified calling interface to achieve unified scheduling of all algorithms in the algorithm library. To ensure the scalability of the algorithm library, each algorithm is self-contained, and it is not required that the data module of a single algorithm can be adapted to other algorithms. Each algorithm exposes its own interface through the interaction part with the unified calling interface. The unified calling interface will detect and call these exposed interfaces to realize the call to each algorithm in the algorithm library. Data modules can only be shared when the inputs of two algorithms in the unified calling interface are the same.

[0032] Specifically, through in-depth research on sentiment analysis algorithms, to enhance their readability, we summarized and categorized them into three core modules: the data module, the model module, and the training module. These three modules essentially correspond to the construction process of each algorithm. The modules of each algorithm are not shared, and module scheduling is not performed at the algorithm library level. Scheduling at the algorithm library level is only possible on an algorithm-by-algorithm basis; that is, the algorithm library level can only see the exposed interface of the algorithm, not each individual module.

[0033] It should be noted that this embodiment summarizes the typical architecture of deep learning algorithms, and organizing algorithms using this architecture can enhance the readability of the algorithm library. However, to enhance scalability, it is not required that all algorithms be organized according to this architecture. Even if some algorithms do not fully possess these three modules, as long as they are self-contained and have an exposed interface 20 to provide a unified calling interface 30 for each algorithm, they can be included in the algorithm library.

[0034] For example, the unified API call can be understood as a Python script. The algorithm name can be specified as a parameter. The unified API call obtains and parses the passed algorithm name parameter, and then, based on this parameter, imports the corresponding algorithm package through the Python standard library importlib. Each algorithm is organized into a Python package and exposed through __init__.py. Therefore, the unified API call can invoke its exposed interface through the imported algorithm package.

[0035] In one implementation, the data module of this embodiment includes:

[0036] The system includes a data download submodule, a data processing submodule, and a dataset construction submodule. The data module processes the input data from the algorithm library and outputs the processed data to the corresponding module.

[0037] In one implementation, the data download submodule described in this embodiment is used to download data, the data processing submodule is used to process all data content related to the algorithm, and the dataset construction submodule is used to construct the dataset interface.

[0038] Specifically, such as Figure 2 As shown, the data module encompasses all data processing related to the algorithm, from initial data download to the construction of the dataset interface and a series of subsequent data-related operations. The processed data can be directly input into the model. To further summarize the functions of the data module, this embodiment divides it into a data download submodule, a data processing submodule, and a dataset construction submodule. Similarly, this division enhances the readability of the algorithm; any algorithm in the loosely coupled sentiment analysis algorithm library may not include a data module.

[0039] In one implementation, the model module of this embodiment includes: a model architecture submodule, a model construction submodule, and a model parameter configuration submodule; the model module is used for model architecture design, algorithm code construction, and parameter configuration.

[0040] In one implementation, the model architecture submodule described in this embodiment is used for model architecture design, the model construction submodule is used for algorithm code construction, and the model parameter configuration submodule is used for configuring model parameters.

[0041] Specifically, such as Figure 2 As shown, the model module is mainly responsible for model architecture design, code construction, and parameter configuration. Here, we separate the model parameter configuration from the unified calling interface, which helps maintain the integrity of each algorithm and makes the unified calling interface more lightweight and applicable to a wider range of situations. To further summarize the functions of the data module, this embodiment divides the model module into a model architecture submodule, a model construction submodule, and a model parameter configuration submodule. Similarly, this division of the model module enhances the readability of the algorithm; any algorithm in the loosely coupled sentiment analysis algorithm library may not contain a model module.

[0042] In one implementation, the training module of this embodiment includes a model training submodule, a model evaluation submodule, and a model inference submodule; the training module is used for model training, evaluation, and inference.

[0043] In one implementation, the model training submodule described in this embodiment is used to train the model, the model evaluation submodule is used to evaluate the training effect of the model, and the model inference submodule is used to optimize the inference of the model.

[0044] Specifically, such as Figure 2 As shown, the training module is mainly responsible for the model's training, evaluation, and inference processes. We separated the model training module from the unified API to ensure the diversity of algorithm training and evaluation metrics. Integrating it into the unified API would significantly limit the scalability of the algorithm package. To further summarize the functions of the data module, this embodiment divides the training module into a model training submodule, a model evaluation submodule, and a model inference submodule to enhance algorithm readability. Furthermore, any algorithm in the loosely coupled sentiment analysis algorithm library can be self-contained and does not include a training module.

[0045] For example, when writing the sentiment analysis algorithm G1, it can be divided into three parts based on its algorithmic logic: a data module, a model module, and a training module. If algorithm G1 does not require data processing at this stage, it can also be divided into two parts: a model module and a training module. Simultaneously, the training module needs to handle both model training and inference; therefore, the algorithm should be written according to the division into training sub-modules and model inference sub-modules. This helps developers quickly understand the algorithm library using a familiar approach, reducing the learning curve and the difficulty of algorithm maintenance.

[0046] In one implementation, the unified invocation interface described in this embodiment is used to obtain invocation parameters, and based on the invocation parameters and the invocation logic, to uniformly invoke any algorithm through the exposed interface of each algorithm in the loosely coupled sentiment analysis algorithm library.

[0047] Specifically, such as Figure 1 As shown, in this embodiment, to achieve unified scheduling of algorithms in the algorithm library, we designed a loosely coupled unified calling interface. This ensures the independence of each algorithm in the library while also enabling unified management and improving the usability of the algorithm library. This loosely coupled unified calling interface also brings the advantage of easy extensibility to the algorithm library. As long as the algorithm can run independently, it can be integrated into the algorithm library by exposing the interface. In the loosely coupled sentiment analysis algorithm library, instead of abstracting a unified module from the perspective of a model or algorithm library, we reduce the coupling between different algorithms by preserving the integrity of each algorithm, allowing each algorithm to be self-contained while achieving unified scheduling through a unified calling interface.

[0048] Specifically, the unified API retrieves the call parameters, which include the algorithm name and algorithm input parameters. The algorithm name can be used to match a specific algorithm in the loosely coupled sentiment analysis algorithm library. Based on the algorithm input parameters and the calling logic, the exposed API of any algorithm can be invoked, and the algorithm can then be called through the exposed API.

[0049] In one implementation, the unified calling interface described in this embodiment is used to search for the hyperparameters of any algorithm in the loosely coupled sentiment analysis algorithm library based on the hyperparameter configuration information.

[0050] Specifically, this embodiment integrates the hyperparameters of all algorithms, and the unified calling interface also supports functions such as hyperparameter search for any algorithm in the algorithm library. First, each hyperparameter is sampled within a reasonable range, then all sample combinations of hyperparameters are traversed, and the unified calling interface is called for training for each hyperparameter combination, thus realizing the hyperparameter search function.

[0051] In one implementation, the loosely coupled sentiment analysis algorithm library described in this embodiment is built on the Ascend framework.

[0052] The MindSpore framework is an open AI framework built through community collaboration to create optimal Ascend matching for all scenarios and support multi-processor architectures. It provides algorithm engineers, data scientists, and developers with a user-friendly, efficient, and flexible deployment experience, helping the domestic AI software and hardware application ecosystem to flourish.

[0053] Specifically, the loosely coupled sentiment analysis algorithm library based on the Ascend framework provided in this embodiment is based on the domestic deep learning framework Ascend, and the Ascend framework itself can support both NPU and GPU chips. It can help developers transition seamlessly from NVIDIA graphics processing units (GPUs) to Huawei Ascend processors (NPUs), achieving full domestic support from hardware to framework.

[0054] In one implementation, the loosely coupled sentiment analysis algorithm library described in this embodiment supports multiple computing platforms, including NPU, GPU, and CPU.

[0055] Specifically, leveraging the cross-platform capabilities of the Ascend framework, the unified API calls support switching between hardware platforms such as NPU, GPU, and CPU via parameters.

[0056] For example, cross-platform compatibility is supported by the Ascend framework, which can be achieved by calling its set_context interface; it should be noted that due to the instability of operators on the NPU, it is sometimes necessary to reimplement specific operators in order to avoid unstable operators.

[0057] In summary, we propose a loosely coupled sentiment analysis algorithm library based on the Ascend framework. This library supports cross-platform compatibility, a unified architecture, a unified API, and easy extensibility. Thanks to the cross-platform capabilities of the domestically developed deep learning framework Ascend, our library can run simultaneously on NVIDIA GPUs and Huawei Ascend NPUs, achieving complete domestic support from hardware to framework. Through research on the architectures of numerous sentiment analysis algorithms, we abstract a unified architecture for sentiment analysis algorithms and build the library based on this unified architecture to improve readability. Furthermore, to achieve unified scheduling of algorithms within the library, we designed a loosely coupled unified API. This ensures the independence of each algorithm while enabling unified management, improving the library's usability. This loosely coupled unified API also brings the advantage of easy extensibility; as long as an algorithm can run independently, it can be integrated into the library by exposing its interface.

[0058] Based on the above embodiments, the present invention also provides a smart terminal, the principle block diagram of which can be as follows: Figure 3 As shown, the smart terminal includes a processor, memory, network interface, display screen, and temperature sensor connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides the environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The network interface is used to communicate with external terminals via a network connection. When executed by the processor, the computer program implements a loosely coupled sentiment analysis algorithm library based on the Ascend framework. The display screen can be an LCD screen or an e-ink screen. The temperature sensor is pre-installed inside the smart terminal to detect the operating temperature of internal devices.

[0059] Those skilled in the art will understand that Figure 3 The block diagram shown is merely a partial structural diagram related to the present invention and does not constitute a limitation on the smart terminal to which the present invention is applied. A specific smart terminal may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0060] In summary, this invention discloses a loosely coupled sentiment analysis algorithm library based on the Ascend framework. The library includes: algorithm modules, exposed interfaces coupled to the algorithm modules, and a unified calling interface coupled to the exposed interfaces. The algorithm modules are used to construct and train algorithms within the library. Each algorithm module includes a data module, a model module, and a training module. The exposed interfaces provide a calling interface for each algorithm to the unified calling interface. The unified calling interface includes hyperparameter configuration information and calling logic. This invention builds the algorithm library based on the domestic Ascend framework, reducing coupling between different algorithms by preserving the integrity of each algorithm. Each algorithm is self-contained while achieving unified scheduling through an exposed interface, thus realizing the library's cross-platform compatibility, unified architecture, unified calling interface, and easy extensibility.

[0061] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A loosely coupled sentiment analysis algorithm library based on the Ascend framework, characterized in that, The loosely coupled sentiment analysis algorithm library includes: an algorithm module, an exposed interface coupled to the algorithm module, and a unified calling interface coupled to the exposed interface. The unified calling interface is used to search for hyperparameters of any algorithm in the loosely coupled sentiment analysis algorithm library based on hyperparameter configuration information, sample each hyperparameter within a reasonable range, traverse all sample combinations of hyperparameters, and call the unified calling interface to train each hyperparameter combination. The algorithm module is used to build and train the algorithms in the loosely coupled sentiment analysis algorithm library. The loosely coupled sentiment analysis algorithm library is built on the Ascend framework and supports multiple computing platforms, including NPU, GPU, and C. The PU, leveraging the cross-platform capabilities of the Ascend framework, provides a unified API that supports switching between NPU, GPU, and CPU hardware platforms via parameters. The algorithm module includes a data module, a model module, and a training module. The exposed interface provides a calling interface for each algorithm within the unified API. This unified API includes hyperparameter configuration information and calling logic. It retrieves calling parameters, including the algorithm name and input parameters, and, based on these parameters and the calling logic, invokes any algorithm through its exposed interface within the loosely coupled sentiment analysis algorithm library. The API calls any algorithm's exposed interface based on the input parameters and calling logic, and then invokes the algorithm itself through the exposed interface.

2. The loosely coupled sentiment analysis algorithm library based on the Ascend framework according to claim 1, characterized in that, The data module includes a data download submodule, a data processing submodule, and a dataset construction submodule; the data module is used to process the data input from the algorithm library and output the processed data to the corresponding module.

3. The loosely coupled sentiment analysis algorithm library based on the Ascend framework according to claim 2, characterized in that, The data download submodule is used to download data, the data processing submodule is used to process all data content related to the algorithm, and the dataset construction submodule is used to build the dataset interface.

4. The loosely coupled sentiment analysis algorithm library based on the Ascend framework according to claim 1, characterized in that, The model module includes: a model architecture submodule, a model construction submodule, and a model parameter configuration submodule; the model module is used for model architecture design, algorithm code construction, and parameter configuration.

5. The loosely coupled sentiment analysis algorithm library based on the Ascend framework according to claim 4, characterized in that, The model architecture submodule is used for model architecture design, the model construction submodule is used for algorithm code construction, and the model parameter configuration submodule is used for configuring model parameters.

6. The loosely coupled sentiment analysis algorithm library based on the Ascend framework according to claim 1, characterized in that, The training module includes a model training submodule, a model evaluation submodule, and a model inference submodule; the training module is used for model training, evaluation, and inference.

7. The loosely coupled sentiment analysis algorithm library based on the Ascend framework according to claim 6, characterized in that, The model training submodule is used to train the model, the model evaluation submodule is used to evaluate the training effect of the model, and the model inference submodule is used to optimize the inference of the model.

8. A smart terminal, characterized in that, The smart terminal includes a memory, a processor, and the Ascend framework-based low-coupling sentiment analysis algorithm library stored in the memory and capable of running on the processor, wherein the processor runs the Ascend framework-based low-coupling sentiment analysis algorithm library as described in any one of claims 1-7.

Images