Topic scoring method, device, server and computer-readable storage medium

Abstract

An embodiment of the present invention provides a topic scoring method, device, server, and computer-readable storage medium, belonging to the technical field of data processing. The method includes: obtaining a topic text to be scored, preprocessing the topic text to obtain a topic list, using a preset word vector generation model to generate a topic vector for each topic word in the topic list, calculating the smallest sphere containing all the topic vectors based on all the topic vectors, and taking the radius of the smallest sphere as the breadth score. By calculating the smallest sphere, the differences between various topic words can be considered, so as to more objectively reflect the true breadth of the topic.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and in particular, to a topic scoring method, apparatus, server, and computer-readable storage medium. Background Art

[0002] A podcast application is an application software for playing audio or video. Users can not only listen to audio and video programs recorded and uploaded by others, but also upload their own recorded audio and video programs to the podcast for sharing. With the popularization and commercialization of podcast applications, there are more and more podcast topics (i.e., podcast content) on the network. Generally, after the podcast application understands and scores the quality of the podcast content, it can perform searches or recommendations based on the scoring results.

[0003] Currently, usually after extracting the topics or keywords of the topic text of the podcast content, scoring the extracted topics or keywords, and using the scoring result as the quality scoring result of the podcast content. However, this scoring method has the problem that it is difficult to reflect the quality of the topic breadth. Summary of the Invention

[0004] In view of this, an object of the present invention is to provide a topic scoring method, apparatus, server, and computer-readable storage medium, which can improve the problem that the existing topic scoring method is difficult to reflect the quality of the topic breadth.

[0005] In order to achieve the above object, the technical solutions adopted in the embodiments of the present invention are as follows.

[0006] In a first aspect, an embodiment of the present invention provides a topic scoring method, adopting the following technical solution.

[0007] A topic scoring method, the method includes:

[0008] Obtain the topic text to be scored;

[0009] Preprocess the topic text to obtain a topic list, where the topic list includes multiple topic words of the topic text;

[0010] Use a preset word vector generation model to generate a topic vector for each of the topic words in the topic list;

[0011] According to all the topic vectors, calculate the smallest sphere containing all the topic vectors, and use the radius of the smallest sphere as the breadth score.

[0012] Further, the step of calculating the smallest sphere containing all the topic vectors according to all the topic vectors includes:

[0013] Based on all the topic vectors, the minimum enclosing ball algorithm is used to calculate the minimum ball that contains all the topic vectors.

[0014] Further, the step of preprocessing the topic text to obtain a topic list includes:

[0015] Using a preset text classification model, classify the topic text to obtain a first topic classification result;

[0016] Extract keywords from the topic text to obtain a second topic classification result;

[0017] Merge the first topic classification result and the second topic classification result to obtain a topic list.

[0018] Further, the step of extracting keywords from the topic text to obtain a second topic classification result includes:

[0019] Segment the topic text to obtain a plurality of terms;

[0020] Calculate the term frequency of each term, and use the terms with a term frequency greater than a preset value as topic words to obtain a second topic classification result, where the second topic classification result includes topic words.

[0021] Further, the step of obtaining the topic text to be scored includes:

[0022] Obtain an audio file to be scored, and decode the audio file to obtain the topic text to be scored.

[0023] Further, the method further includes the step of obtaining a word vector generation model, and this step includes:

[0024] Obtain a general corpus, segment the general corpus to obtain a plurality of terms;

[0025] Calculate the term frequency of each term, and build a Huffman tree based on the term frequency, where the Huffman tree uses each term as a node and the term frequency as a weight;

[0026] Based on the Huffman tree, perform unsupervised training on the word vector related model to obtain a word vector generation model.

[0027] Further, the step of performing unsupervised training on the word vector related model based on the Huffman tree to obtain a word vector generation model includes:

[0028] Initialize the parameters of the word vector related model;

[0029] Using the word vectors of each of the said terms as inputs and taking the output of the Huffman tree as the target, train the word vector related model to obtain a word vector generation model.

[0030] In a second aspect, the present invention provides a topic scoring device, adopting the following technical solution.

[0031] A topic scoring device includes a receiving and processing module and a scoring module;

[0032] The receiving and processing module is used to obtain the topic text to be scored and to preprocess the topic text to obtain a topic list, where the topic list includes multiple topic words of the topic text;

[0033] The scoring module is used to use a preset word vector generation model to generate topic vectors for each of the topic words in the topic list, calculate the smallest sphere containing all the topic vectors based on all the topic vectors, and use the radius of the smallest sphere as the breadth score.

[0034] In a third aspect, an embodiment of the present invention provides a server, adopting the following technical solution.

[0035] A server includes a processor and a memory, where the memory stores machine-executable instructions that can be executed by the processor, and the processor can execute the machine-executable instructions to implement the topic scoring method as described in the first aspect.

[0036] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium, adopting the following technical solution.

[0037] A computer-readable storage medium has a computer program stored thereon, and when the computer program is executed by a processor, it implements the topic scoring method as described in the first aspect.

[0038] The topic scoring method, device, server, and computer-readable storage medium provided by the embodiments of the present invention preprocess the topic text to be scored to obtain a topic list including multiple topic words, thereby using the word vector generation model to obtain the topic vectors of each topic word, and then calculating the smallest sphere containing all the topic vectors and using the radius of the smallest sphere as the breadth score. By calculating the smallest sphere, the differences between each topic word can be considered, so that the true breadth of the topic can be more objectively reflected, and thus the problem that the existing topic scoring method is difficult to reflect the breadth quality of the topic can be improved.

[0039] To make the above objects, features, and advantages of the present invention more obvious and understandable, the following specific preferred embodiments are given and detailed descriptions are made in conjunction with the accompanying drawings as follows. Description of the Drawings

[0040] To more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the accompanying drawings required for the embodiments. It should be understood that the following drawings only show some embodiments of the present invention, and thus should not be regarded as limiting the scope. For those of ordinary skill in the art, without creative efforts, other related drawings can also be obtained based on these drawings.

[0041] Figure 1 It shows a block diagram of the topic scoring system provided by an embodiment of the present invention.

[0042] Figure 2 It shows Figure 1 a block diagram of the server in

[0043] Figure 3 It shows a flowchart of some steps of the topic scoring method provided by an embodiment of the present invention.

[0044] Figure 4 It shows Figure 3 a flowchart of some sub-steps of step S103 in

[0045] Figure 5 It shows Figure 4 a flowchart of some sub-steps of step S103-2 in

[0046] Figure 6 It shows a flowchart of another part of the steps of the topic scoring method provided by an embodiment of the present invention.

[0047] Figure 7 It shows Figure 6 a flowchart of some sub-steps of step S303 in

[0048] Figure 8 It shows a block diagram of the topic scoring device provided by an embodiment of the present invention.

[0049] Icons: 100 - topic scoring system; 110 - server; 120 - client; 130 - memory; 140 - processor; 150 - communication module; 160 - topic scoring device; 170 - receiving and processing module; 180 - scoring module. Detailed implementation manners

[0050] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, rather than all embodiments. Usually, the components of the embodiments of the present invention described and shown in the drawings here can be arranged and designed in various different configurations.

[0051] Accordingly, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely represents selected embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative efforts fall within the scope of protection of the present invention.

[0052] It should be noted that relational terms such as "first" and "second" are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprising", "including" or any other variant thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements not only includes those elements, but also includes other elements not expressly listed, or elements inherent to such process, method, article or device. Without further limitation, an element defined by the phrase "comprising a..." does not exclude the presence of additional identical elements in the process, method, article or device comprising the element.

[0053] A podcast application is an application software for playing audio or video. Users can not only listen to the audio and video programs recorded and uploaded by others, but also upload their own recorded audio and video programs to the podcast for sharing. With the popularization and commercialization of podcast applications, there are more and more podcast topics (i.e., podcast content) on the Internet. Generally, after the podcast application understands and scores the quality of the podcast content, it can search or recommend according to the scoring results.

[0054] One of the key technologies for podcast content quality scoring is to score the topic breadth of the podcast. At present, most text topic extraction can only extract topics or keywords, resulting in an inability to obtain an accurate topic breadth score. Moreover, generally only the number of topics included is calculated, without considering the differences between each topic, making it difficult for this scoring method to reflect the true breadth quality of the topics.

[0055] Based on the above considerations, the embodiments of the present invention provide a topic scoring scheme, which can improve the problem that the current topic scoring method is difficult to translate the true breadth quality of the topics. Hereinafter, the topic scoring scheme provided by the present invention will be introduced from the perspectives of the topic scoring method and the topic scoring device 160, etc.

[0056] Please refer to Figure 1, which is a block diagram of the topic scoring system 100. The topic scoring method provided by the embodiments of the present invention is applied to the topic scoring system 100. The topic scoring system 100 includes a server 110 and a plurality of clients 120. The server 110 is communicatively connected to the plurality of clients 120 through a network. After the topic text to be scored is sent to the server 110 by the client 120, the server 110 uses the topic scoring method provided by the embodiments of the present invention to perform a breadth score on the topic text to be scored, so as to improve the problem that it is difficult to reflect the breadth quality of the current topic scoring method.

[0057] A podcast sharing platform may be deployed in the server 110. The client 120 may log in to the podcast sharing platform through the network to upload audio and video information to the podcast sharing platform or share the podcast content in the podcast sharing platform.

[0058] Among them, the client 120 may be, but is not limited to, various personal computers, laptop computers, smart phones, tablet computers, and portable wearable devices. The server 110 may be implemented by an independent server 110 or a server cluster composed of a plurality of servers 110.

[0059] Please refer to Figure 2 , which is a block diagram of the server 110. The server 110 includes a memory 130, a processor 140, and a communication module 150. The elements of the memory 130, the processor 140, and the communication module 150 are directly or indirectly electrically connected to each other to realize data transmission or interaction. For example, these elements may be electrically connected to each other through one or more communication buses or signal lines.

[0060] Among them, the memory 130 is used to store programs or data. The memory 130 may be, but is not limited to, a random access memory (RAM), a read only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), etc.

[0061] The processor 140 is configured to read / write data, computer programs, or machine-executable instructions stored in the memory 130 and perform corresponding functions. In this embodiment, the processor 140 reads and executes the computer programs or machine-executable instructions stored in the memory 130 to implement the topic scoring method provided by the present invention.

[0062] The communication module 150 is configured to establish a communication connection between the server 110 and other communication terminals through a network and to send and receive data through the network.

[0063] It should be understood that Figure 2 the structure shown is only a schematic diagram of the structure of the server 110, and the server 110 may further include more or fewer components than those Figure 2 shown, or have a configuration different from that Figure 2 shown. Figure 2 Each component shown can be implemented by hardware, software, or a combination thereof.

[0064] In one embodiment, as Figure 3 shown, a topic scoring method is provided. This embodiment mainly takes the application of this method to the Figure 1 server 110 as an example to illustrate, and includes the following steps.

[0065] S101, obtain the topic text to be scored.

[0066] The client 120 can log in to the podcast sharing platform on the server 110 to upload the topic text to be scored to the server 110, and the server 110 receives the topic text.

[0067] S103, preprocess the topic text to obtain a topic list.

[0068] Among them, the topic list includes multiple topic words of the topic text. The topic words can be keywords, topic words, content words, or main idea words, etc., of the topic text. The multiple topic words can reflect the main content or key content, etc., of the topic text.

[0069] S105, use a preset word vector generation model to generate a topic vector for each topic word in the topic list.

[0070] S107, calculate the smallest sphere containing all the topic vectors according to all the topic vectors, and use the radius of the smallest sphere as the breadth score.

[0071] In the above topic scoring method, by preprocessing the topic text to be scored, a topic list including multiple topic words is obtained, and then using a word vector generation model, the topic vectors of each topic word are obtained. Furthermore, the smallest sphere containing all the topic vectors is calculated, and the radius of the smallest sphere is used as the breadth score. By calculating the smallest sphere, the differences between various topic words can be considered, so that the true breadth of the topic can be more objectively reflected, and thus the problem that the existing topic scoring method is difficult to reflect the breadth quality of the topic can be improved.

[0072] It should be understood that the smallest sphere is the smallest sphere containing all the topic words. Therefore, the smallest sphere can consider the differences between various topic words and then reflect the true breadth of the topic text to be scored.

[0073] Regarding S101, the method for obtaining the topic text to be scored can be flexibly selected. For example, the client 120 can directly send the topic text to be scored to the server 110, or the server 110 can extract the topic text to be scored or the podcast audio from the podcast application software. In one implementation, obtaining the topic text to be scored may include: obtaining the audio file to be scored and decoding the audio file to obtain the topic text to be scored.

[0074] Furthermore, after the server 110 obtains the audio file to be scored, the audio file can be decoded using speech recognition technology or audio decoding technology to obtain the topic text to be scored.

[0075] After obtaining the topic text, the server 110 can flexibly select the method for preprocessing the topic text. For example, a trained model can be used to extract the topic words, or the topic text can be segmented to obtain the topic words.

[0076] In one implementation, referring to Figure 4 , Figure 4 is a schematic flowchart of some sub-steps of the above step S103. The preprocessing of the topic text is realized through the following sub-steps to obtain a topic list.

[0077] S103-1, using a preset text classification model to classify the topic text to obtain the first topic classification result.

[0078] S103-2, extracting keywords from the topic text to obtain the second topic classification result.

[0079] S103-3, combining the first topic classification result and the second topic classification result to obtain a topic list.

[0080] Through the above steps S103-1 to S103-3, using the text classification model, classify the topic text to obtain the first topic classification result. The topic words in the first topic classification result are rough topics, that is, topic words with relatively rough disassembling and analyzing. Extract keywords from the topic text to obtain the second topic classification result. The topic words in the second topic classification result are fine topics, that is, topic words with more detailed disassembling and analyzing (the splitting granularity is finer than that of the first topic classification result).

[0081] Finally, merge the first topic classification result and the second topic classification result. The obtained topic list includes both rough topic words and fine topic words, so that the topic list can contain various semantics of the topic text as much as possible, and further helps to more truly reflect the topic breadth.

[0082] Among them, the text classification model can be selected according to actual needs, and can be but not limited to: text classification models such as textcnn model, fasttext model, TextRNN model, etc.

[0083] Furthermore, for S103-2, the way of extracting keywords can be flexibly selected. For example, keyword extraction technology can be used to extract keywords from the topic text, the TF-IDF algorithm can be used to extract keywords from the topic text, or neural networks can be used to extract keywords.

[0084] In one implementation, referring to Figure 5 , which is the process schematic diagram of some sub-steps of S103-2, extract keywords from the topic text through the following steps to obtain the second topic classification result.

[0085] S201, segment the topic text to obtain multiple word items.

[0086] Use the word segmentation method to segment the topic text to obtain multiple word items. Among them, the word segmentation method can be selected according to needs. For example, it can be but not limited to: neural networks, dictionary-based word segmentation algorithms, and statistical machine learning algorithms.

[0087] S202, calculate the word frequency of each word item, and use the word items with word frequency greater than the preset value as topic words to obtain the second topic classification result.

[0088] Among them, the second topic classification result includes topic words, and the preset value can include word frequency value or ranking.

[0089] When the preset value is the word frequency value, use the word items with word frequency greater than the word frequency value as topic words.

[0090] When the preset value is the ranking, sort the word items according to the word frequency, and select the word items within the ranking range as topic words.

[0091] For example, if the ranking is among the top 5, then the terms ranked among the top 5 are used as topic terms.

[0092] Furthermore, the topic scoring method provided by the embodiments of the present invention further includes the step of obtaining a word vector generation model. Refer to Figure 6 , which is a schematic flowchart of partial sub-steps of this step, including the following steps.

[0093] S301: Obtain a general corpus, perform word segmentation on the general corpus to obtain a plurality of terms.

[0094] The general corpus can be the collected corpus or the corpus obtained from the general corpus database.

[0095] S302: Calculate the word frequency of each term, and build a Huffman tree based on the word frequency.

[0096] Among them, the Huffman tree uses each term as a node and the word frequency of the term as the weight.

[0097] The word frequency from the root node to the child nodes of the Huffman tree can decrease layer by layer, that is, the deeper the child node, the smaller its word frequency.

[0098] S303: Based on the Huffman tree, perform unsupervised training on the word vector related model to obtain a word vector generation model.

[0099] Among them, the word vector related model can be selected but not limited to training methods such as the Word2Vec algorithm and the Glove algorithm for training.

[0100] In one implementation, the word vector related model selects the Word2Vec algorithm for training. On this basis, refer to Figure 7 , which is a schematic flowchart of partial sub-steps of the above S303, including the following steps.

[0101] S303-1: Initialize the parameters of the word vector related model.

[0102] S303-2: Use the word vectors of each term as input, and use the output Huffman tree as the target to train the word vector related model to obtain a word vector generation model.

[0103] Among them, the word vectors can be stored in the batch database. Based on this, S303-2 can include the following steps: Take out data from the batch database as input; perform forward propagation to obtain the value of the likelihood function; use the gradient descent method to update the model parameters. Loop the above steps until the model converges.

[0104] Therefore, by adopting the steps, a word vector generation model is trained by combining the general corpus and the Word2Vec algorithm.

[0105] Furthermore, for S107, the method for calculating the minimum sphere can be flexibly selected. In one implementation, the minimum sphere containing all topic vectors can be calculated through the following steps: Based on all topic vectors, the minimum sphere covering algorithm is used to calculate the minimum sphere containing all topic vectors.

[0106] Optionally, the minimum covering sphere algorithm may include the following steps.

[0107] S1. Sort all topic vectors, and let the minimum covering sphere of the first i - 1 topic vectors be C.

[0108] S2. Determine whether the i-th topic vector is within C. If so, execute S3. Otherwise, execute S4.

[0109] S3. Determine that the minimum covering sphere of the first i topic vectors is also C.

[0110] S4. Determine the minimum covering sphere C1 of the first i topic vectors.

[0111] Among them, S4 may include the following method: Let the current center of the sphere be Pi and the radius be 0, and fix the i-th topic vector (i.e., the topic vector Pi) as a point; find another topic vector Pj that is not on the current minimum covering sphere, let the current center of the sphere be (Pi + Pj) / 2, the radius be ∣PiPj∣ / 2, and find the minimum sphere covering of the j-th topic vector Pj and the i-th topic vector Pi; if there is no minimum sphere on which both the topic vector Pj and the topic vector Pi are on the circle, continue to find a topic vector Pk that is not on the current minimum sphere, and let the current minimum covering sphere be the circumscribed sphere of the topic vectors Pi, Pj, and Pk to find the minimum covering sphere of the topic vectors Pi, Pj, and Pk.

[0112] It should be understood that after finding three topic vectors, according to the principle of determining a circle by three points, the minimum covering sphere on which all three topic vectors are on the circle can definitely be obtained.

[0113] In this way, by iterating the value of i and repeating S2 - S4, the minimum covering sphere containing all topic vectors can be determined.

[0114] The topic scoring method provided by the embodiments of the present invention coarsely classifies the topic text to be scored by using a text classification model to obtain a first topic classification result, and extracts keywords from the topic text according to the word frequency, that is, performs fine classification to obtain a second topic classification result. After fusing the first topic classification result and the second topic classification result, a topic list is obtained, so that the topic list can contain various semantic information of the topic text as much as possible. Furthermore, a preset word vector generation model is used to obtain topic vectors for each topic word in the topic list, and the smallest sphere containing all the topic vectors is obtained, and the radius of the smallest sphere is used as the breadth score, which can consider the differences between each topic word and objectively reflect the true breadth of the topic.

[0115] It should be understood that although Figures 3 - 7 the steps in the flowchart of Figures 3 - 7 are shown in sequence according to the indication of the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless there is a clear description in this article, the execution of these steps has no strict order limit, and these steps can be executed in other orders. Moreover,

[0116] To execute the corresponding steps in the above embodiments and each possible manner, an implementation manner of a topic scoring device 160 is given below. Optionally, the topic scoring device 160 can adopt the device structure of the server 110 shown in the above Figure 2 Further, please refer to Figure 8 Figure 8 which is a functional module diagram of a topic scoring device 160 provided by the embodiments of the present invention. It should be noted that the basic principle and the technical effects generated by the topic scoring device 160 provided in this embodiment are the same as those in the above embodiments. For a brief description, for the parts not mentioned in this embodiment, reference can be made to the corresponding content in the above embodiments. The topic scoring device 160 includes a receiving and processing module 170 and a scoring module 180.

[0117] The receiving and processing module 170 is used to obtain the topic text to be scored and preprocess the topic text to obtain a topic list.

[0118] Among them, the topic list includes multiple topic words of the topic text.

[0119] ​The scoring module 180 is used to generate a topic vector for each topic word in the topic list using a preset word vector generation model, calculate the minimum sphere containing all topic vectors based on all topic vectors, and use the radius of the minimum sphere as the breadth score.

[0120] Optionally, the above modules can be stored in the form of software or firmware. Figure 2 The memory 130 shown in FIG. 130 or the operating system (OS) of the server 110 may be fixed and may be used by Figure 2 Meanwhile, the data, program codes, etc. required to execute the above modules may be stored in the memory 130.

[0121] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can also be implemented in other ways. The device embodiments described above are merely illustrative. For example, the flowcharts and block diagrams in the accompanying drawings show the possible architectures, functions and operations of the devices, methods and computer program products according to multiple embodiments of the present invention. In this regard, each box in the flowchart or block diagram can represent a module, a program segment or a portion of code, and the module, program segment or a portion of code contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in an order different from that marked in the accompanying drawings. For example, two consecutive boxes can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flowchart, and the combination of boxes in the block diagram and / or flowchart, can be implemented using a dedicated hardware-based system that performs the specified function or action, or can be implemented using a combination of dedicated hardware and computer instructions.

[0122] In addition, the functional modules in the various embodiments of the present invention may be integrated together to form an independent part, or each module may exist independently, or two or more modules may be integrated to form an independent part.

[0123] When the above-mentioned function is implemented in the form of a software functional module and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical discs.

[0124] The foregoing are only the preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and changes. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A topic scoring method, characterized in that, The method includes: Obtain the topic text to be scored; Preprocess the topic text to obtain a topic list, where the topic list includes multiple topic words of the topic text; Use a preset word vector generation model to generate topic vectors for each of the topic words in the topic list; According to all the topic vectors, calculate the minimum sphere containing all the topic vectors, and use the radius of the minimum sphere as the breadth score; The step of preprocessing the topic text to obtain a topic list includes: Use a preset text classification model to classify the topic text to obtain a first topic classification result; Extract keywords from the topic text to obtain a second topic classification result; Merge the first topic classification result and the second topic classification result to obtain a topic list.

2. The topic scoring method according to claim 1, characterized in that, The step of calculating the minimum sphere containing all the topic vectors according to all the topic vectors includes: Based on all the topic vectors, use the minimum sphere covering algorithm to calculate the minimum sphere containing all the topic vectors.

3. The topic scoring method according to claim 1, wherein The step of extracting keywords from the topic text to obtain a second topic classification result includes: Segment the topic text to obtain multiple word items; Calculate the word frequency of each word item, and use the word items with word frequencies greater than a preset value as topic words to obtain a second topic classification result, where the second topic classification result includes topic words.

4. The topic scoring method according to claim 1, characterized in that, The step of obtaining the topic text to be scored includes: Obtain an audio file to be scored, and decode the audio file to obtain the topic text to be scored.

5. The topic scoring method according to any one of claims 1 to 4, characterized in that, The method further includes the step of obtaining a word vector generation model, and this step includes: Obtain a general corpus, segment the general corpus to obtain multiple word items; Calculate the word frequency of each word item, and establish a Huffman tree according to the word frequency, where the Huffman tree uses each word item as a node and the word frequency as a weight; Based on the Huffman tree, perform unsupervised training on a word vector related model to obtain a word vector generation model.

6. The topic scoring method according to claim 5, characterized in that The step of performing unsupervised training on a word vector related model based on the Huffman tree to obtain a word vector generation model includes: Initialize the parameters of the word vector related model; Use the word vectors of each word item as input, and use the output Huffman tree as the target to train the word vector related model to obtain a word vector generation model.

7. A topic scoring device, characterized in that, It includes a receiving and processing module and a scoring module; The receiving and processing module is used to obtain the topic text to be scored, and is also used to preprocess the topic text to obtain a topic list, where the topic list includes multiple topic words of the topic text; The scoring module is used to use a preset word vector generation model to generate topic vectors for each of the topic words in the topic list, calculate the minimum sphere containing all the topic vectors according to all the topic vectors, and use the radius of the minimum sphere as the breadth score; The receiving and processing module is further used for: Use a preset text classification model to classify the topic text to obtain a first topic classification result; Extract keywords from the topic text to obtain a second topic classification result; Merge the first topic classification result and the second topic classification result to obtain a topic list.

8. A server, characterized in that, It includes a processor and a memory. The memory stores machine-executable instructions that can be executed by the processor, and the processor can execute the machine-executable instructions to implement the topic scoring method according to any one of claims 1-6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the topic scoring method according to any one of claims 1-6.

Images