Model training, title extraction method, device, electronic device and computer readable medium

Abstract

The embodiment of the present invention provides a model training, title extraction method, device, electronic device and computer-readable medium. The model training method includes: obtaining a training sample, the training sample includes a plurality of SMS content sentences and the correlation information between the plurality of SMS content sentences and SMS titles; using at least the plurality of SMS content sentences as input, and at least the correlation information as the first output, training a neural network to obtain an SMS title extraction model. In the scheme of the embodiment of the present invention, since the SMS title extraction model is obtained by training with a plurality of SMS content sentences and the correlation information between the plurality of SMS content sentences and SMS titles as training samples, the SMS title extraction model can be used to extract SMS content sentences as SMS titles, thereby improving the efficiency of SMS title extraction while ensuring the reliability of SMS title extraction, and further improving the effect of SMS display.

Description

Technical Field

[0001] Embodiments of the present invention relate to the field of computer technology, and in particular, to a model training, title extraction method, device, electronic device and computer-readable medium. Background Art

[0002] In mobile terminals, services that convert text messages such as notifications and verification codes into card messages are gradually emerging. For example, a mobile terminal can be installed with SDK (Software Development Kit) software for SMS services. When the mobile terminal obtains notifications and verification codes, the SDK software can process them based on various scenarios to obtain card messages, which is more conducive to showing the scene content to users.

[0003] However, for marketing scenarios, existing technologies cannot display SMS content that meets user expectations. Summary of the invention

[0004] In view of this, an embodiment of the present invention provides a model training, title extraction method, device, electronic device and computer-readable medium to solve or alleviate the above problems.

[0005] According to a first aspect of an embodiment of the present invention, a model training method is provided, comprising: obtaining a training sample, the training sample comprising a plurality of SMS content sentences and correlation information between the plurality of SMS content sentences and SMS titles; training a neural network using at least the plurality of SMS content sentences as input and at least the correlation information as a first output to obtain a SMS title extraction model.

[0006] According to a second aspect of an embodiment of the present invention, a title extraction method is provided, comprising: inputting a plurality of SMS content sentences of a title SMS to be extracted into an SMS title extraction model to obtain title extraction information of the plurality of SMS content sentences, wherein the SMS title extraction model is trained by the training method described in the first aspect; and determining the title of the title SMS to be extracted based on the title extraction information.

[0007] According to a third aspect of an embodiment of the present invention, a model training device is provided, comprising: a sample acquisition module, which acquires training samples, wherein the training samples include multiple SMS content sentences and correlation information between the multiple SMS content sentences and SMS titles; a model training module, which uses at least the multiple SMS content sentences as input and at least the correlation information as a first output to train a neural network to obtain an SMS title extraction model.

[0008] According to a fourth aspect of an embodiment of the present invention, a title extraction device is provided, comprising: a model processing module, inputting a plurality of SMS content sentences of a title SMS to be extracted into an SMS title extraction model, and obtaining title extraction information of the plurality of SMS content sentences, wherein the SMS title extraction model is obtained by training according to the training method described in the first aspect; and a title determination module, determining the title of the title SMS to be extracted based on the title extraction information.

[0009] According to a fifth aspect of an embodiment of the present invention, there is provided an electronic device, comprising: a processor, a memory, a communication interface and a communication bus, wherein the processor, the memory and the communication interface communicate with each other via the communication bus; the memory is used to store at least one executable instruction, and the executable instruction enables the processor to perform an operation corresponding to the method described in the first aspect or the second aspect.

[0010] According to a sixth aspect of an embodiment of the present invention, there is provided a computer-readable medium on which a computer program is stored, and when the program is executed by a processor, the method as described in the first aspect or the second aspect is implemented.

[0011] In the scheme of the embodiment of the present invention, since the SMS title extraction model is trained by using multiple SMS content sentences and the correlation information between multiple SMS content sentences and SMS titles as training samples, the SMS title extraction model can extract SMS content sentences as SMS titles, thereby ensuring the reliability of SMS title extraction while improving the efficiency of SMS title extraction, and further improving the effect of SMS display. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the embodiments of the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0013] Figure 1 A schematic diagram of a method for processing text message data according to an embodiment of the present invention;

[0014] Figure 2A A schematic flow chart of a model training method according to another embodiment of the present invention;

[0015] Figure 2B A schematic diagram of a model training method according to another embodiment of the present invention;

[0016] Figure 3A A schematic diagram of a model training method according to another embodiment of the present invention;

[0017] Figure 3B A schematic diagram of a model training method according to another embodiment of the present invention;

[0018] Figure 4A A schematic flow chart of a model training method according to another embodiment of the present invention;

[0019] Figure 4B A schematic diagram of a model training method according to another embodiment of the present invention;

[0020] Figure 5A is a schematic flow chart of a title extraction method according to another embodiment of the present invention;

[0021] Figure 5B is a schematic diagram of a title extraction method according to another embodiment of the present invention;

[0022] Figure 6 is a schematic block diagram of a model training device according to another embodiment of the present invention;

[0023] Figure 7 is a schematic block diagram of a title extraction device according to another embodiment of the present invention;

[0024] Figure 8 The present invention is another embodiment of the hardware structure of an electronic device. DETAILED DESCRIPTION

[0025] In order to enable those skilled in the art to better understand the technical solutions in the embodiments of the present invention, the technical solutions 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. All other embodiments obtained by ordinary technicians in the field based on the embodiments in the embodiments of the present invention should fall within the scope of protection of the embodiments of the present invention.

[0026] The specific implementation of the embodiment of the present invention is further described below in conjunction with the accompanying drawings of the embodiment of the present invention. Figure 1Schematic diagram of a method for processing SMS data according to an embodiment of the present invention. As shown in the figure, two-way communication can be achieved between a SMS operation service end 120 such as operated by a mobile network operator, a mobile terminal 110 such as a mobile phone, and a SMS scene service end 130 such as operated by a SMS scene service provider. The SMS operation service end 120 and the mobile terminal 110 can communicate through a mobile communication network such as 2G, 3G, 4G or 5G. The mobile terminal 110 and the SMS scene service end 130 can communicate through a network such as the mobile Internet. The above communication data includes but is not limited to voice, text, pictures or streaming media data. As shown in the figure, in this example, the mobile terminal 110 includes a mobile network module 111, an Internet module 112 and a presentation module 113, but it should be understood that in other examples, the mobile terminal 110 may also include other modules. Among them, the mobile network module 111 can be implemented as a communication device such as a transmitter, a receiver or a transceiver, and a Subscriber Identity Module (SIM). The Internet module 112 can be implemented as a network adapter, etc. Among them, the above modules are only exemplary. For example, the mobile network module 111 and the Internet module 112 can be implemented as a single device or include a common part. The presentation module 113 can be implemented by a human-computer interaction interface component such as a display or a touch screen, and can be configured accordingly through the user space of the operating system. For example, a text message application can be configured in the operating system, and the text message application can be configured to receive text messages or send text messages. The text message application can also be configured with data processing functions such as rendering processing through SDK (software development kit). For example, the text message application can also communicate with the text message scene server 130 to realize the scene function of the text message. For example, the text message scene server 130 can be implemented as an intelligent text message platform based on semantic precision recognition technology, which reconstructs visual data by identifying the text message content and provides third-party services to users according to different scenarios. The text message application can be used as the service distribution entrance closest to user needs, providing users with special functions such as text message public accounts, scenario pop-ups, rich bubbles, and smart card holders. The SMS application can also be combined with any application that can obtain mobile phone text messages (for example, social applications, search applications, e-commerce applications, media applications, etc.) to achieve smoother user services.

[0027] For SMS marketing scenarios, the length of SMS content ranges from dozens to hundreds of words. Marketing SMS is SMS for marketing purposes, and usually contains discounts, red envelopes, big promotions, and full-reduction activities. The large amount of information and the wide variety of information make it difficult for users to read comprehensively and effectively, and valuable information is difficult to be delivered, thus failing to efficiently display SMS content that meets user expectations.

[0028] Figure 2A A schematic flowchart of a model training method according to another embodiment of the present invention. Figure 2A The model training method can be applied to any appropriate electronic device with data processing capability, including but not limited to: a server, a mobile terminal (such as a mobile phone, a PAD, etc.) and a PC, etc. The method comprises:

[0029] 210: Obtain training samples, where the training samples include a plurality of SMS content sentences and correlation information between the plurality of SMS content sentences and SMS titles.

[0030] It should be understood that the SMS content sentences can be obtained by sentence processing the SMS content data. The multiple SMS content sentences can be continuous sentences in the SMS content data, or can be multiple discrete sentences, which is not limited in the embodiment of the present invention. The SMS content data can be text data including characters such as letters or Chinese characters, numbers, etc. The SMS content data in the text can be complete SMS data, that is, data extracted from a single SMS file. The SMS content data can also be data extracted from a part of the SMS file, for example, it can be data of SMS content sentences. The above SMS content data can be represented by word vectors. For example, after obtaining the training sample, the obtained text data can be represented in the form of word vectors (word embedding). The data represented by the word vector can also be obtained as a training sample. The SMS content data can include all SMS content data, for example, content data obtained from the SMS file. The SMS content data can also be SMS content data after preprocessing, for example, in the content data obtained from the SMS file, irrelevant information such as signatures, short links, and unsubscribe information is removed to obtain SMS content data as a training sample. The SMS type labels can be manually labeled based on experience or experimental results, or the SMS data content can be labeled by a neural network labeling model, or the above model can be used for auxiliary labeling.

[0031] It should also be understood that the relevance information of multiple SMS content sentences and SMS titles can be implemented in any form. For example, the relevance information can be represented in digital form (for example, a one-dimensional vector), or in vector form, that is, a relevance information vector. The relevance information vector can be a probability vector including multiple dimensions, and the multiple dimensions correspond to multiple SMS content sentences. Preferably, the correspondence is one-to-one. Each dimension of the probability vector can indicate the relevance of the SMS content sentence to the title of the SMS file (the SMS file from which the SMS content data is extracted). Preferably, the sum of the multiple annotated dimension values ​​corresponding to the multiple dimensions is 1. The multiple annotated dimension values ​​can also be other numerical values ​​and are normalized.

[0032] In one example, the multiple SMS content sentences are consecutive sentences in the same SMS file. For example, when the number of SMS content sentences is 4, [1, 0, 0, 0] may indicate that the first SMS content sentence is marked as a title.

[0033] In another example, multiple SMS content sentences are consecutive sentences in the same SMS file. For example, when the number of SMS content sentences is 4, [0.1, 0.5, 0.2, 0.2] may indicate that all four sentences have the possibility of being called titles, but the second sentence has the greatest possibility of being a title. This is because the criteria for selecting SMS titles may be complex and diverse. The above-mentioned annotation processing may be performed manually based on experience values ​​or experimental values, or may be assisted by a neural network, which is not limited in the embodiments of the present invention.

[0034] In another example, the multiple SMS content sentences are non-continuous sentences in the same SMS file. For example, when the number of SMS content sentences is 3, [1,0,0] may indicate that the first SMS content sentence is marked as a title.

[0035] In another example, the plurality of SMS content sentences are non-continuous sentences in different SMS files. For example, when the number of SMS content sentences is 5, [0, 0, 0, 0, 1] may indicate that the fifth SMS content sentence is marked as a title.

[0036] 220: Taking at least a plurality of SMS content sentences as input and at least relevance information as a first output, training a neural network to obtain an SMS title extraction model.

[0037] It should be understood that the training sample includes multiple groups of sub-training samples, and the relevance information vectors of different sub-training samples may have different dimensions. The relevance information vectors with the same dimension and the multiple SMS content sentences corresponding to the number of dimensions may be trained together (for example, in the same batch or in different batches). Different sub-training samples may be trained in different batches. Thus, the diversity of samples is achieved and the training model is obtained.

[0038] It should also be understood that the above-mentioned neural networks may include convolutional neural networks (CNN), recurrent neural networks (RNN) and other neural networks.

[0039] In the scheme of the embodiment of the present invention, since the SMS title extraction model is trained by using multiple SMS content sentences and the correlation information between multiple SMS content sentences and SMS titles as training samples, the SMS title extraction model can extract SMS content sentences as SMS titles, thereby ensuring the reliability of SMS title extraction while improving the efficiency of SMS title extraction, and further improving the effect of SMS display.

[0040] In addition, since the SMS content is divided into sentences and the titles are annotated for training, the obtained model realizes end-to-end title extraction processing.

[0041] Deep learning applications or artificial intelligence (AI) applications (i.e., so-called intelligent applications hereinafter) can be implemented using one computing engine or multiple computing engines. Applications may include data engines, preprocessing engines, model reasoning engines, and post-processing engines, etc. It should be understood that in other examples, other engines may also be included. For marketing scenarios, the data engine can be used to obtain data from a network interface of a mobile terminal such as a mobile phone. In this example, the preprocessing engine can be used to directly call methods such as network interfaces to implement text preprocessing capabilities. In addition, the model reasoning engine can be used to call the interface provided by the deep learning model management module to load offline model files and complete the reasoning calculation of the model. In addition, the post-processing engine can be used to complete the post-processing of the output data of the model reasoning.

[0042] The above process enables the SMS title extraction task in the application. Each engine can complete its own function relatively independently. For example, the preprocessing engine can be used to complete the preprocessing function using a text preprocessing module (such as SMS content sentence processing, or word vector embedding processing). In addition, the model inference engine can be used to load and execute offline models. In addition, the calculation engine can be customized by developers.

[0043] In addition, when developing applications, one or more application development or program development can be performed based on the number of applications running simultaneously on the hardware platform. In addition, each application can enable multiple threads to complete reasoning tasks. Each application can also start a single thread reasoning task.

[0044] In addition, for offline models, it can be divided into the model generation stage, the application compilation stage, and the application deployment stage. In the model generation stage, the model can be parsed and converted into (for example) a customized intermediate representation of a computational graph. In the development environment, kernel fusion is performed through an offline model generator, and after processing such as operator optimization, an offline model raw file is generated. The offline model can be directly deployed in the operating system in the production environment.

[0045] Figure 2B It is a schematic diagram of a model training method of another embodiment of the present invention. As shown in the figure, in this example, three SMS content sentences are taken as examples. SMS content sentence 1, SMS content sentence 2 and SMS content sentence 3 are used as input, and the three title probabilities (correlation information with the SMS title) title probability 1, title probability 2 and title probability 3 are the corresponding annotation information for learning of the three sentences. For example, the training sample SMS is "The shopping carnival is coming, the domestic hot-selling TV is directly reduced by 50%, and it will be launched at 11 o'clock tonight", and the corresponding three sentences are "The shopping carnival is coming". "The domestic hot-selling TV is directly reduced by 50%." "It will be launched at 11 o'clock tonight". For example, for marketing scenarios, the most suitable sentence among these three sentences as the SMS title is sentence 2, "The domestic hot-selling TV is directly reduced by 50%". Therefore, in an example, the probability of the second sentence can be set to 1, and the probabilities of sentences 1 and 3 can be set to 0. In another example, although sentence 2 is the most suitable as a text message title, the other sentences still have certain information transmission value in the marketing scenario, so the probability of sentence 2 can be set to 0.8, and the probabilities of sentences 1 and 3 can be set (marked as) 0.1.

[0046] It should be understood that the above-mentioned annotation can be performed by manual annotation, for example, by industry experience or by statistical experimental results. The above-mentioned annotation can also be performed by an annotation model obtained by machine learning. For example, a relevant information prediction model of a classification model, a clustering model, etc. can be obtained by machine learning, and a specific sentence is input into the model to obtain the probability of annotation. In addition, the above-mentioned two annotation methods can also be combined.

[0047] In another implementation of the present invention, the training sample also includes sequence information between multiple SMS content sentences, and the neural network includes a first convolutional neural network and a second convolutional neural network, wherein at least multiple SMS content sentences are used as input, and at least relevance information is used as the first output to train the neural network, including: using multiple SMS content sentences as inputs of the first convolutional neural network respectively, and using parallel processing results of multiple outputs of the first convolutional neural network based on sequence information as inputs of the second convolutional neural network, and using at least the relevance information as the first output to train the first convolutional neural network and the second convolutional neural network.

[0048] Since the input of the second convolutional neural network is the parallel processing result of multiple outputs of the first convolutional neural network based on sequence information, the global features of multiple sentences can be efficiently learned through the second convolutional neural network, and the input of the first convolutional neural network is multiple SMS content sentences, so the local features of each sentence can be efficiently learned through the second convolutional neural network, thereby improving the stability of the SMS title extraction model. In addition, since convolutional neural networks can be used for good parallel data processing, convolutional neural networks are used for training to improve the data processing efficiency in model training.

[0049] In another implementation of the present invention, the training samples also include title-related label information of each of the multiple SMS content sentences, wherein the neural network is trained using at least the multiple SMS content sentences as input and at least the relevance information as the first output, including: training the neural network using at least the multiple SMS content sentences as input, the relevance information as the first output, and the title-related label information as the second output.

[0050] Since title-related label information is an efficient representation when extracting titles, it is annotated with relevance information and title-related label information for supervised learning, thereby adopting the above-mentioned multi-task training to improve the training efficiency of the model.

[0051] In another implementation of the present invention, the training sample also includes sequence information between multiple SMS content sentences, and the neural network includes a first convolutional neural network and a second convolutional neural network, wherein at least multiple SMS content sentences are used as input, relevance information is used as a first output, and title-related label information is used as a second output to train the neural network, including: using multiple SMS content sentences as inputs of the first convolutional neural network respectively, and using parallel processing results of multiple outputs of the first convolutional neural network based on sequence information as inputs of the second convolutional neural network, and at least using relevance information as a first output and title-related label information as a second output to train the first convolutional neural network and the second convolutional neural network.

[0052] In this example, based on the above reasons, the above-mentioned multi-task training is adopted to improve the training efficiency of the model; the convolutional neural network is used for training to improve the data processing efficiency in model training; and the stability of the SMS title extraction model is improved through global feature learning and local special effect learning.

[0053] In another implementation of the present invention, the training sample also includes sentence length information of each of the multiple SMS content sentences, wherein the parallel processing results of the multiple outputs of the first convolutional neural network based on the sequence information are used as the input of the second convolutional neural network, including: adding the sentence length information of each of the multiple SMS content sentences to the multiple outputs of the first convolutional neural network respectively, and processing the multiple outputs in parallel to obtain the input of the second convolutional neural network.

[0054] Since the length of the sentence has a great influence on the result of title extraction when extracting the SMS content sentence as the title extraction, the input of the second convolutional neural network includes the sentence length information of multiple SMS content sentences, thereby improving the model training efficiency.

[0055] In another implementation of the present invention, obtaining training samples includes: classifying a plurality of SMS content sentences based on at least one title-related tag to obtain respective title-related tag information.

[0056] Since the respective title-related tag information can be obtained through at least one title-related tag, manual tagging is avoided, thereby improving tagging efficiency and accuracy.

[0057] In another implementation of the present invention, the title-related label information is a title-related label vector, at least one dimension of the related label vector corresponds to at least one title-related label, and the at least one title-related label includes at least one of an interest point, a triggering point, a product description, and an activity description.

[0058] Since interest points, triggering points, product descriptions and activity descriptions are the most important content features in marketing scenarios, using at least one of the above features as at least one dimension of the relevant label vector improves the efficiency of model training and obtains a more reliable model.

[0059] For example, interest points may include information that is of direct benefit to users, such as text descriptions of discounts, red envelopes, and discounts. For example, "50% off". Trigger points may include content used to build momentum for marketing activities, usually including time information. For example, text descriptions such as "Start buying at 9 o'clock tonight". Product categories may include text descriptions of a certain type of product, that is, product name information, such as "domestic mobile phones", "imported cars", etc. Activity descriptions may include descriptive copy of a certain type of marketing activity, which usually does not include specific interest points, such as "Queen's Day is coming", "Shopping Carnival", etc.

[0060] It should be understood that the marketing value of the above-mentioned various title-related tags can be set by users based on experience or experiments. Generally, the marketing value of information on interest points and triggering points is higher, and the marketing value is positively correlated with the dimensional value of the title-related tags. Accordingly, the dimensional value of interest points and triggering points can be set higher than other dimensional values.

[0061] In another implementation of the present invention, obtaining training samples includes: obtaining multiple SMS content sentences by sentence processing on content data of a target SMS, wherein the multiple SMS content sentences include a title sentence and a non-title sentence as a title of the SMS, the relevance information of the title sentence is a first probability, the relevance information of the non-title sentence is a second probability, and the first probability is greater than the second probability.

[0062] Since the relevance information of the title sentence is marked as the first probability and the relevance information of the non-label sentence is marked as the second probability, efficient and convenient marking of the relevance information is achieved.

[0063] In one example, the multiple SMS content sentences include a single title sentence and multiple non-title sentences (other SMS content sentences), and the sum of the probabilities of the multiple SMS content sentences is 1. The first probability is 1, and the second probability is 0.

[0064] In another example, multiple SMS content sentences include multiple title sentences and at least one non-title sentence, and the sum of the probabilities of the multiple SMS content sentences is 1. The first probabilities of the multiple title sentences may be the same or different. The second probability of at least one non-title sentence is less than the first probability. When at least one non-title sentence is a plurality of non-title sentences, the second probabilities may be the same or different. For example, multiple SMS content sentences may include two title sentences and two non-title sentences, wherein the probabilities of the two title sentences are 0.4 respectively; and the probabilities of the two non-title sentences are 0.1 respectively. For another example, the probabilities of the two title sentences are 0.4 and 0.3 respectively; and the probabilities of the two non-title sentences are 0.1 and 0.2 respectively.

[0065] Figure 3A A schematic diagram of a model training method according to another embodiment of the present invention. Figure 3A The model training method can be applied to any appropriate electronic device with data processing capability, including but not limited to: a server, a mobile terminal (such as a mobile phone, a PAD, etc.) and a PC, etc. The method comprises:

[0066] 310: Acquire training samples, where the training samples include a plurality of SMS content sentences, relevance information between the plurality of SMS content sentences and SMS titles, and title-related label information of the plurality of SMS content sentences.

[0067] 320: Taking at least a plurality of SMS content sentences as input, taking relevance information as a first output, and taking title related tag information as a second output, training a neural network to obtain an SMS title extraction model.

[0068] In the solution of the embodiment of the present invention, since the title-related label information is an efficient representation when extracting the title, the relevance information and the title-related label information are respectively annotated for supervised learning, thereby improving the training efficiency of the model.

[0069] In another implementation of the present invention, obtaining training samples includes: obtaining multiple SMS content sentences by sentence processing on content data of a target SMS, wherein the multiple SMS content sentences include a title sentence and a non-title sentence as a title of the SMS, the relevance information of the title sentence is a first probability, the relevance information of the non-title sentence is a second probability, and the first probability is greater than the second probability.

[0070] In another implementation of the present invention, obtaining training samples includes: classifying a plurality of SMS content sentences based on at least one title-related tag to obtain respective title-related tag information.

[0071] In another implementation of the present invention, the title-related tag information is a title-related tag vector, at least one dimension of the tag vector corresponds to at least one title-related tag, and the at least one title-related tag includes at least one of an interest point, a trigger point, a product description, and an activity description.

[0072] Figure 3B It is a schematic diagram of a model training method of another embodiment of the present invention. As shown in the figure, in this example, SMS content sentence 1, SMS content sentence 2 and SMS content sentence 3 are respectively used as inputs, and accordingly, title probability 1, title probability 2 and title probability 3 are respectively used as first outputs, and accordingly, sentence label 1, sentence label 2 and sentence label 3 are respectively used as second outputs. Sentence labels can have multiple dimensions, for example, including three dimensions of triggering point, interest point and recent activity. For example, the labels can be represented as vectors [A, B, C], where A, B and C represent triggering point, interest point and recent activity respectively. For example, sentence label 1, sentence label 2 and sentence label 3 can be [1, 0, 0], [0, 1, 0] and [1, 0, 1] respectively. It can be seen that the main attribute of sentence label 1 is the triggering point, the main attribute of sentence label 2 is the interest point, and the main attribute of sentence label 3 is the triggering point and recent activity. It should be understood that the elements in the above vectors can be integers or fractions, and the embodiment of the present invention is not limited to this.

[0073] Figure 4A A schematic flowchart of a model training method according to another embodiment of the present invention. Figure 4A The model training method can be applied to any appropriate electronic device with data processing capability, including but not limited to: a server, a mobile terminal (such as a mobile phone, a PAD, etc.) and a PC, etc. The method comprises:

[0074] 410: Obtain training samples, where the training samples include a plurality of SMS content sentences, correlation information between the plurality of SMS content sentences and SMS titles, and sequence information between the plurality of SMS content sentences.

[0075] 420: Using multiple SMS content sentences as inputs of a first convolutional neural network, and using multiple outputs of the first convolutional neural network as parallel processing results based on sequence information as inputs of a second convolutional neural network, and using at least relevance information as a first output, the first convolutional neural network and the second convolutional neural network are trained to obtain a SMS title extraction model.

[0076] Since the input of the second convolutional neural network is the parallel processing result of multiple outputs of the first convolutional neural network based on sequence information, the global features of multiple sentences can be efficiently learned through the second convolutional neural network, and the input of the first convolutional neural network is multiple SMS content sentences, so the local features of each sentence can be efficiently learned through the second convolutional neural network, thereby improving the stability of the SMS title extraction model. In addition, since convolutional neural networks can be used for good parallel data processing, convolutional neural networks are used for training to improve the data processing efficiency in model training.

[0077] In another implementation of the present invention, obtaining training samples includes: obtaining multiple SMS content sentences by sentence processing on content data of a target SMS, wherein the multiple SMS content sentences include a title sentence and a non-title sentence as a title of the SMS, the relevance information of the title sentence is a first probability, the relevance information of the non-title sentence is a second probability, and the first probability is greater than the second probability.

[0078] Figure 4BIt is a schematic diagram of a model training method of another embodiment of the present invention. As shown in the figure, in this example, three text message content sentences are taken as examples. Text message content sentence 1, text message content sentence 2 and text message content sentence 3 are respectively used as inputs of local convolutional neural network 1, local convolutional neural network 2 and local convolutional neural network 3. The output of local convolutional neural network 1, local convolutional neural network 2 and local convolutional neural network 3 can be the output vector obtained by processing through their respective pooling layers, and the three output vectors can be constructed as an input matrix as the input of the global convolutional neural network, and the three title probabilities (correlation information with the text message title) title probability 1, title probability 2 and title probability 3 are the corresponding annotation information for learning of the three sentences, as the output of the global convolutional neural network. In addition, in one example, the sentence length addition processing can be performed before the input matrix is ​​input into the global convolutional neural network. For example, the sentence length information of the sentence vector can be added to the input matrix corresponding to different sentence vectors. In another example, the sentence length information of the sentence vector can be added after the output vector obtained by processing through the respective pooling layers, and then, multiple vectors with sentence length information added are constructed as input matrices. In another example, the sentence length information may also be added to the input matrix of the corresponding local convolutional neural network (the sentence matrix formed by word vectors representing the sentences). This embodiment of the present invention is not limited to this.

[0079] The local convolutional neural network and the global convolutional neural network can be the same convolutional neural network or different convolutional neural networks. For example, the two can have the same or different convolutional kernel sizes, the same or different convolutional kernel expansion coefficients, the same or different numbers of convolutional layers, etc.

[0080] Figure 5A A schematic flowchart of a title extraction method according to another embodiment of the present invention. Figure 5A The title extraction method can be applied to any appropriate electronic device with data processing capability, including but not limited to: a server, a mobile terminal (such as a mobile phone, a PAD, etc.) and a PC, etc. The method comprises:

[0081] 510: Input multiple SMS content sentences of the SMS with titles to be extracted into the SMS title extraction model to obtain title extraction information of the multiple SMS content sentences.

[0082] 520: Based on the title extraction information, determine the title of the title SMS to be extracted.

[0083] In the scheme of the embodiment of the present invention, since the SMS title extraction model is trained by using multiple SMS content sentences and the correlation information between multiple SMS content sentences and SMS titles as training samples, the SMS title extraction model can extract SMS content sentences as SMS titles, thereby ensuring the reliability of SMS title extraction while improving the efficiency of SMS title extraction, and further improving the effect of SMS display.

[0084] In addition, since the SMS content is divided into sentences and the titles are annotated for training, the obtained model realizes end-to-end title extraction processing.

[0085] In one example, a first terminal (e.g., a mobile terminal or a PC terminal) can obtain a text message title extraction model from a server terminal. The first terminal (e.g., a mobile terminal or a PC terminal) can receive a text message with a title to be extracted from a second terminal (e.g., via a server terminal). The first terminal can (e.g., via a server terminal) feedback the title of the text message with a title to be extracted to the second terminal. For example, the first terminal can be a user terminal (C terminal), and the second terminal can be a merchant (B terminal). In addition, the second terminal can send a revised title of the text message with a title to be extracted to the server terminal. The server terminal can train the text message title extraction model based on the revised title of the text message with a title to be extracted, so as to update the text message title extraction model, and the first terminal can obtain the updated text message title extraction model from the server terminal. Thus, efficient dynamic update of model training and model deployment is achieved through the above-mentioned interactive process.

[0086] In another example, a text message generation model or a text message generation template can be generated on the server side, and the server side can deploy the text message generation model or the text message generation template to the second terminal. Among them, the text message title extraction model can generate a target title based on the input target text message. The text message generation model can generate a target text message based on the input target title. The text message generation template can assist the second terminal in generating a target text message based on the target title. Since the text message generation model or the text message generation template of the second terminal corresponds to and matches the text message title extraction model of the first terminal, a standardized and structured marketing text message format is provided for the merchant, and the model prediction accuracy of the user side is improved.

[0087] Figure 5B This is a schematic diagram of a title extraction method according to another embodiment of the present invention. As shown in the figure, the SMS content data can be input into the SMS sentence processing module to obtain SMS content sentence 1, SMS content sentence 2, and SMS content sentence 3 (different from the above-mentioned training process, the sentences here are the sentences input when the model is used for reasoning). The SMS extraction model obtained by the above training is used to perform reasoning and prediction on each sentence to obtain the title probability corresponding to each sentence. In one example, the sentence corresponding to the maximum probability can be directly determined as the sentence serving as the SMS title. In another example, Figure 3B Corresponding to the model in , the title-related label vector of each sentence can be obtained, and the title-related label vector can be used as a reference to assist in judging the sentence as the SMS title. In other words, the sentence as the SMS title can be judged based on both the size of the probability value and the validity of the title-related label vector. In some cases, the title-related label vector of the sentence corresponding to the maximum probability may not be judged to be valid. In this case, the sentence may not be the most suitable as the SMS title. As a result, the prediction reliability of the model is improved.

[0088] In another implementation of the present invention, the method further includes: performing sentence processing on the title SMS to be extracted to obtain multiple SMS content sentences. Since the sentence processing can effectively obtain each sentence of the title SMS to be extracted, and each sentence can be used as input in the model for efficient title extraction, the complexity of the model is reduced while ensuring data processing efficiency.

[0089] In one example, the text message content sentence with the highest probability can be selected from multiple probabilities as the title of the text message to be extracted. For example, the content of the text message can be "Queen's Day is coming, well-known brand handbags are directly reduced by 50%, don't miss it, it will be available at 9 o'clock tonight." The vector of the title extraction information of the corresponding multiple sentences is represented as [0.1, 0.4, 0.3, 0.2], where each dimension corresponds to the first sentence, the second sentence, the third sentence and the fourth sentence respectively. The second sentence corresponding to the probability of 0.4 ("well-known brand handbags are directly reduced by 50%) can be extracted as the text message title and displayed to the user accordingly.

[0090] In another implementation of the present invention, the title extraction information includes multiple probabilities that multiple SMS content sentences are the titles of the title SMS to be extracted, and title-related label information of each of the multiple SMS content sentences, wherein the title of the title SMS to be extracted is determined based on the title extraction information, including: if the title-related label information of each of the multiple SMS content sentences is judged to be valid, then based on the multiple probabilities, the title of the title SMS to be extracted is determined.

[0091] Since the title extraction information is utilized based on the validity of the title-related tag information, the reliability of title extraction using the model is improved.

[0092] Figure 6 The present invention is a schematic block diagram of a model training device according to another embodiment of the present invention. Figure 6 The model training device can be applied to any appropriate electronic device with data processing capability, including but not limited to: a server, a mobile terminal (such as a mobile phone, a PAD, etc.) and a PC, etc. The device includes:

[0093] The sample acquisition module 610 acquires a training sample, where the training sample includes a plurality of SMS content sentences and correlation information between the plurality of SMS content sentences and SMS titles.

[0094] The model training module 620 takes at least a plurality of SMS content sentences as input and at least relevance information as a first output, trains a neural network, and obtains an SMS title extraction model.

[0095] In the scheme of the embodiment of the present invention, since the SMS title extraction model is trained by using multiple SMS content sentences and the correlation information between multiple SMS content sentences and SMS titles as training samples, the SMS title extraction model can extract SMS content sentences as SMS titles, thereby ensuring the reliability of SMS title extraction while improving the efficiency of SMS title extraction, and further improving the effect of SMS display.

[0096] In another implementation of the present invention, the training sample also includes sequence information between multiple SMS content sentences, and the neural network includes a first convolutional neural network and a second convolutional neural network, wherein the model training module is specifically used to: use the multiple SMS content sentences as inputs of the first convolutional neural network, and use the parallel processing results of the multiple outputs of the first convolutional neural network based on the sequence information as inputs of the second convolutional neural network, and use at least the relevance information as the first output to train the first convolutional neural network and the second convolutional neural network.

[0097] In another implementation of the present invention, the training samples also include title-related label information of each of the multiple SMS content sentences, wherein the model training module is specifically used to train the neural network using at least multiple SMS content sentences as input, relevance information as the first output, and title-related label information as the second output.

[0098] In another implementation of the present invention, the training sample also includes sequence information between multiple SMS content sentences, and the neural network includes a first convolutional neural network and a second convolutional neural network, wherein the model training module is specifically used to: use the multiple SMS content sentences as inputs of the first convolutional neural network, and use the parallel processing results of the multiple outputs of the first convolutional neural network based on the sequence information as inputs of the second convolutional neural network, and at least use the relevance information as the first output and the title-related label information as the second output to train the first convolutional neural network and the second convolutional neural network.

[0099] In another implementation of the present invention, the training sample also includes sentence length information of each of the multiple SMS content sentences, wherein the model training module is specifically used to: obtain the input of the second convolutional neural network by adding the sentence length information of each of the multiple SMS content sentences to the multiple outputs of the first convolutional neural network, and performing parallel processing on the multiple outputs.

[0100] In another implementation of the present invention, the sample acquisition module is specifically used to: classify a plurality of SMS content sentences based on at least one title-related tag to obtain respective title-related tag information.

[0101] In another implementation of the present invention, the title-related label information is a title-related label vector, at least one dimension of the related label vector corresponds to at least one title-related label, and the at least one title-related label includes at least one of an interest point, a triggering point, a product description, and an activity description.

[0102] In another implementation of the present invention, the sample acquisition module is specifically used to: obtain multiple SMS content sentences by sentence processing on the content data of the target SMS, wherein the multiple SMS content sentences include a title sentence and a non-title sentence serving as the title of the SMS, the relevance information of the title sentence is a first probability, the relevance information of the non-title sentence is a second probability, and the first probability is greater than the second probability.

[0103] The device of this embodiment is used to implement the corresponding methods in the aforementioned multiple method embodiments, and has the beneficial effects of the corresponding method embodiments, which will not be repeated here. In addition, the functional implementation of each module in the device of this embodiment can refer to the description of the corresponding parts in the aforementioned method embodiments, which will not be repeated here.

[0104] Figure 7 It is a schematic block diagram of a title extraction device according to another embodiment of the present invention. Figure 7 The title extraction device can be applied to any appropriate electronic device with data processing capability, including but not limited to: a server, a mobile terminal (such as a mobile phone, a PAD, etc.) and a PC, etc. The device includes:

[0105] The model processing module 710 inputs a plurality of SMS content sentences of the SMS with titles to be extracted into the SMS title extraction model to obtain title extraction information of the plurality of SMS content sentences.

[0106] The title determination module 720 determines the title of the SMS message to be extracted based on the title extraction information.

[0107] In the scheme of the embodiment of the present invention, since the SMS title extraction model is trained by using multiple SMS content sentences and the correlation information between multiple SMS content sentences and SMS titles as training samples, the SMS title extraction model can extract SMS content sentences as SMS titles, thereby ensuring the reliability of SMS title extraction while improving the efficiency of SMS title extraction, and further improving the effect of SMS display.

[0108] In another implementation of the present invention, the device further includes a sentence segmentation module for performing sentence segmentation processing on the title SMS to be extracted to obtain a plurality of SMS content sentences.

[0109] In another implementation of the present invention, the title extraction information includes multiple probabilities that multiple SMS content sentences are the titles of the title SMS to be extracted, and title-related label information of each of the multiple SMS content sentences, wherein the title determination module is specifically used to: if the title-related label information of each of the multiple SMS content sentences is judged to be valid, then determine the title of the title SMS to be extracted based on multiple probabilities.

[0110] The device of this embodiment is used to implement the corresponding methods in the aforementioned multiple method embodiments, and has the beneficial effects of the corresponding method embodiments, which will not be repeated here. In addition, the functional implementation of each module in the device of this embodiment can refer to the description of the corresponding parts in the aforementioned method embodiments, which will not be repeated here.

[0111] Figure 8 The hardware structure of the electronic device of another embodiment of the present invention is as follows; Figure 8 As shown, the hardware structure of the electronic device may include: a processor 801, a communication interface 802, a computer readable medium 803 and a communication bus 804;

[0112] The processor 801, the communication interface 802, and the computer readable medium 803 communicate with each other via a communication bus 804;

[0113] Optionally, the communication interface 802 may be an interface of a communication module;

[0114] The processor 801 may be specifically configured to: obtain a training sample, the training sample comprising a plurality of SMS content sentences and relevance information between the plurality of SMS content sentences and SMS titles; train a neural network using at least the plurality of SMS content sentences as input and at least the relevance information as a first output to obtain a SMS title extraction model;

[0115] Alternatively, multiple SMS content sentences of the SMS with titles to be extracted are input into an SMS title extraction model to obtain title extraction information of the multiple SMS content sentences; based on the title extraction information, the title of the SMS with titles to be extracted is determined.

[0116] The processor 801 may be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it may also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components. It may implement or execute the methods, steps and logic block diagrams disclosed in the embodiments of the present invention. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0117] The computer readable medium 803 can be, but is not limited to, a random access memory medium (RAM), a read-only memory medium (ROM), a programmable read-only memory medium (PROM), an erasable read-only memory medium (EPROM), an electrically erasable read-only memory medium (Electric Erasable Programmable Read-Only Memory, EEPROM), etc.

[0118] In particular, according to an embodiment of the present disclosure, the process described above with reference to the flowchart can be implemented as a computer software program. For example, an embodiment of the present disclosure includes a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program includes a program code configured to execute the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from the network through a communication part, and / or installed from a removable medium. When the computer program is executed by a central processing unit (CPU), the above functions defined in the method of the present invention are executed. It should be noted that the computer-readable medium of the present invention can be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two. The computer-readable medium can be, for example, but not limited to, a system, device or device of electricity, magnetism, light, electromagnetic, infrared, or semiconductor, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to, an electrical connection with one or more conductors, a portable computer disk, a hard disk, a random access storage medium (RAM), a read-only storage medium (ROM), an erasable programmable read-only storage medium (EPROM or flash memory), an optical fiber, a portable compact disk read-only storage medium (CD-ROM), an optical storage medium, a magnetic storage medium, or any suitable combination of the above. In the present invention, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in combination with an instruction execution system, an apparatus or a device. In the present invention, a computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, in which a computer-readable program code is carried. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. A computer-readable signal medium may also be any computer-readable medium other than a computer-readable storage medium, which may send, propagate, or transmit a program configured to be used by or in combination with an instruction execution system, an apparatus or a device. The program code embodied on the computer readable medium may be transmitted using any appropriate medium, including but not limited to wireless, wireline, optical fiber cable, RF, etc., or any suitable combination of the foregoing.

[0119] Computer program code configured to perform the operations of the present invention may be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, C++, and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., through the Internet using an Internet service provider).

[0120] The flowcharts and block diagrams in the accompanying drawings illustrate the possible architecture, functions and operations of the systems, methods and computer program products according to various 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 part of a code, and the module, program segment, or a part of the code contains one or more executable instructions configured to implement the specified logical function. There are specific sequential relationships in the above-mentioned specific embodiments, but these sequential relationships are only exemplary. When the specific implementation is implemented, these steps may be less, more or the execution order may be adjusted. That is, in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two boxes represented in succession 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 with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0121] The modules involved in the embodiments of the present invention may be implemented by software or hardware. The names of these modules do not limit the modules themselves in some cases.

[0122] As another aspect, the present invention further provides a computer-readable medium having a computer program stored thereon, which implements the method described in the above embodiment when the program is executed by a processor.

[0123] As another aspect, the present invention also provides a computer-readable medium, which may be included in the device described in the above embodiment; or may exist independently without being assembled into the device. The above computer-readable medium carries one or more programs, and when the above one or more programs are executed by the device, the device: obtains training samples, the training samples include multiple SMS content sentences and correlation information between the multiple SMS content sentences and SMS titles; uses at least the multiple SMS content sentences as input and at least the correlation information as the first output to train a neural network to obtain an SMS title extraction model;

[0124] Alternatively, multiple SMS content sentences of the SMS with titles to be extracted are input into an SMS title extraction model to obtain title extraction information of the multiple SMS content sentences; based on the title extraction information, the title of the SMS with titles to be extracted is determined.

[0125] The expressions "first", "second", "first" or "second" used in various embodiments of the present disclosure may modify various components regardless of order and / or importance, but these expressions do not limit the corresponding components. The above expressions are only configured for the purpose of distinguishing an element from other elements. For example, a first user device and a second user device represent different user devices, although both are user devices. For example, a first element may be referred to as a second element, and similarly, a second element may be referred to as a first element without departing from the scope of the present disclosure.

[0126] When one element (e.g., a first element) is referred to as being "(operably or communicatively) coupled" or "(operably or communicatively) coupled to" or "connected to" another element (e.g., a second element), it is understood that the one element is directly connected to the other element or the one element is indirectly connected to the other element via yet another element (e.g., a third element). Conversely, it is understood that when an element (e.g., a first element) is referred to as being "directly connected" or "directly coupled" to another element (the second element), no element (e.g., a third element) is interposed between the two.

[0127] The above description is only a preferred embodiment of the present invention and an explanation of the technical principles used. Those skilled in the art should understand that the scope of the invention involved in the present invention is not limited to the technical solution formed by a specific combination of the above technical features, but should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the above inventive concept. For example, the above features are replaced with the technical features with similar functions disclosed in the present invention (but not limited to) to form a technical solution.

Claims

1. A model training method, comprising: Acquire a training sample, wherein the training sample includes a plurality of SMS content sentences and correlation information between the plurality of SMS content sentences and SMS titles; Using at least the multiple SMS content sentences as input and at least the relevance information as a first output, training a neural network to obtain an SMS title extraction model; The training sample further includes the sentence length information of each of the multiple SMS content sentences, and the multiple SMS content sentences and the sentence length information of each of the multiple SMS content sentences are used as input to train the neural network to obtain the SMS title extraction model; The obtaining of training samples includes: performing sentence processing on content data of a target SMS to obtain the plurality of SMS content sentences, wherein the plurality of SMS content sentences include a title sentence and a non-title sentence serving as a title of the SMS, the relevance information of the title sentence is a first probability, the relevance information of the non-title sentence is a second probability, and the first probability is greater than the second probability.

2. The method according to claim 1, wherein: The training sample also includes sequence information between the multiple text message content sentences, the neural network includes a first convolutional neural network and a second convolutional neural network, wherein at least taking the multiple text message content sentences as input and at least taking the relevance information as a first output, training the neural network includes: The multiple SMS content sentences are respectively used as inputs of the first convolutional neural network, and the multiple outputs of the first convolutional neural network are used as inputs of the second convolutional neural network based on the parallel processing results of the sequence information, and at least the relevance information is used as the first output to train the first convolutional neural network and the second convolutional neural network.

3. The method according to claim 1, wherein: The training sample also includes title-related label information of each of the plurality of SMS content sentences, wherein the step of taking at least the plurality of SMS content sentences as input and at least the relevance information as a first output to train the neural network includes: The neural network is trained by at least taking the multiple SMS content sentences as input, taking the relevance information as a first output, and taking the title related tag information as a second output.

4. The method according to claim 3, wherein: The training sample also includes sequence information between the multiple text message content sentences, the neural network includes a first convolutional neural network and a second convolutional neural network, wherein at least taking the multiple text message content sentences as input, taking the relevance information as a first output, and taking the title-related tag information as a second output, training the neural network includes: The multiple SMS content sentences are used as inputs of the first convolutional neural network respectively, and the multiple outputs of the first convolutional neural network are used as inputs of the second convolutional neural network based on the parallel processing results of the sequence information, and at least the relevance information is used as the first output and the title-related tag information is used as the second output to train the first convolutional neural network and the second convolutional neural network.

5. The method according to claim 4, wherein: The method of using the parallel processing results of the multiple outputs of the first convolutional neural network based on the sequence information as the input of the second convolutional neural network includes: The input of the second convolutional neural network is obtained by respectively adding the sentence length information of each of the multiple SMS content sentences to the multiple outputs of the first convolutional neural network and processing the multiple outputs in parallel.

6. The method according to claim 3, wherein: The obtaining of training samples comprises: Based on at least one title-related tag, the multiple SMS content sentences are classified and processed to obtain respective title-related tag information.

7. The method according to claim 6, wherein: The title-related tag information is a title-related tag vector, at least one dimension of the related tag vector corresponds to the at least one title-related tag, and the at least one title-related tag includes at least one of an interest point, a trigger point, a product description, and an activity description.

8. A title extraction method, comprising: Inputting multiple SMS content sentences of the SMS to be extracted into the SMS title extraction model to obtain title extraction information of the multiple SMS content sentences, wherein the SMS title extraction model is trained by the training method according to claims 1-7; Based on the title extraction information, the title of the title SMS to be extracted is determined.

9. The method according to claim 8, wherein: The method further comprises: The title SMS to be extracted is processed into sentences to obtain the multiple SMS content sentences.

10. The method according to claim 8, wherein: The title extraction information includes multiple probabilities that the multiple SMS content sentences are the titles of the SMS to be extracted, and title-related tag information of each of the multiple SMS content sentences, wherein: The step of determining the title of the SMS message to be extracted based on the title extraction information includes: If the title-related tag information of each of the plurality of SMS content sentences is determined to be valid, the title of the SMS to be extracted is determined based on the plurality of probabilities.

11. A model training device, comprising: A sample acquisition module, which acquires a training sample, wherein the training sample includes a plurality of SMS content sentences and correlation information between the plurality of SMS content sentences and SMS titles; A model training module, which uses at least the multiple SMS content sentences as input and at least the relevance information as a first output to train a neural network to obtain an SMS title extraction model; The training sample also includes the sentence length information of each of the multiple SMS content sentences, and the model training module is further used to take the multiple SMS content sentences and the sentence length information of each of the multiple SMS content sentences as input to train the neural network to obtain the SMS title extraction model; The sample acquisition module acquires training samples, including: obtaining the multiple SMS content sentences by sentence processing on the content data of the target SMS, wherein the multiple SMS content sentences include a title sentence and a non-title sentence as the title of the SMS, the relevance information of the title sentence is a first probability, the relevance information of the non-title sentence is a second probability, and the first probability is greater than the second probability.

12. A title extraction device, comprising: A model processing module, inputting a plurality of SMS content sentences of the SMS to be extracted into an SMS title extraction model, and obtaining title extraction information of the plurality of SMS content sentences, wherein the SMS title extraction model is trained by the training method according to claims 1-7; The title determination module determines the title of the SMS message to be extracted based on the title extraction information.

13. An electronic device, comprising: A processor, a memory, a communication interface and a communication bus, wherein the processor, the memory and the communication interface communicate with each other via the communication bus; The memory is used to store at least one executable instruction, and the executable instruction enables the processor to perform an operation corresponding to the method according to any one of claims 1 to 10.

14. A computer readable medium having a computer program stored thereon, wherein when the program is executed by a processor, the method according to any one of claims 1 to 10 is implemented.

Images