Contextual online advertisement serving method, apparatus, server, and storage medium

Abstract

The present disclosure relates to a context online advertisement delivery method and device, a server and a storage medium, and relates to the technical field of computers. The method comprises: obtaining a text to be processed; in response to a user performing a context operation on the text to be processed, obtaining a target text unit and a target text, obtaining a word document corresponding table and a candidate sentence library; obtaining first target advertisement information and second target advertisement information according to the target text unit, the word document corresponding table, the target text and the candidate sentence library, providing the target text and / or the target text unit, the first target advertisement information and / or the second target advertisement information to the user. Thus, the user can establish a global view of the language structure and train the language sense to quickly master the language. At the same time, the user is provided with advertisement information related to the learning text, which improves the click rate of the advertisement while realizing the implantation of the advertisement scene.

Description

Technical Field

[0001] This disclosure relates to the field of computer technology, and in particular to a method, apparatus, server, and storage medium for contextual online advertising. Background Technology

[0002] In related technologies, users learn languages ​​through some applications. However, these language learning applications can only provide translations or pronunciations of sentences, and users can only learn the meaning or pronunciation of sentences through these applications. Furthermore, these language learning applications are only used for language learning and have limited functionality. Summary of the Invention

[0003] This disclosure provides a contextual online advertising delivery method, apparatus, server, and storage medium to at least address the problem of limited functionality in language learning applications in related technologies. The technical solution of this disclosure is as follows:

[0004] According to a first aspect of the present disclosure, a contextual online advertising delivery method is provided, comprising: acquiring text to be processed; wherein the text to be processed includes a plurality of text units, the text units being words or phrases; in response to a user performing a contextual operation on the text to be processed, acquiring target text units and target text; acquiring a word-document mapping table and a candidate statement library; wherein the word-document mapping table includes a plurality of candidate words and advertising information associated with the candidate words, and the candidate statement library includes a plurality of candidate statements and advertising information associated with the candidate statements; acquiring first target advertising information based on the target text units and the word-document mapping table; acquiring second target advertising information based on the target text and the candidate statement library; and providing the user with the target text and / or the target text units, as well as the first target advertising information and / or the second target advertising information.

[0005] According to a second aspect of the present disclosure, a context-based online advertising delivery device is provided, comprising: a text acquisition unit for acquiring text to be processed; wherein the text to be processed includes a plurality of text units, the text units being words or phrases; a target acquisition unit for acquiring target text units and target text in response to a user performing a context operation on the text to be processed; a data acquisition unit for acquiring a word-document correspondence table and a candidate statement library; wherein the word-document correspondence table includes a plurality of candidate words and advertising information associated with the candidate words, and the candidate statement library includes a plurality of candidate statements and advertising information associated with the candidate statements; a first information acquisition unit for acquiring first target advertising information based on the target text units and the word-document correspondence table; a second information acquisition unit for acquiring second target advertising information based on the target text and the candidate statement library; and an information providing unit for providing the user with the target text and / or the target text units, as well as the first target advertising information and / or the second target advertising information.

[0006] According to a third aspect of the present disclosure, an electronic device is provided, comprising: a processor; and a memory for storing processor-executable instructions; wherein the processor is configured to execute the instructions to implement the contextual online advertising delivery method as described in the first aspect above.

[0007] According to a fourth aspect of the present disclosure, a storage medium is provided that, when instructions in the storage medium are executed by a processor of an electronic device, enables the electronic device to perform the contextual online advertising delivery method as described in the first aspect above.

[0008] According to a fifth aspect of the present disclosure, a computer program product is provided, including a computer program that, when executed by a processor, implements the contextual online advertising delivery method as described in the first aspect above.

[0009] The technical solutions provided by the embodiments of this disclosure bring at least the following beneficial effects:

[0010] By implementing the embodiments of this disclosure, a text to be processed is obtained; in response to a user performing contextual operations on the text to be processed, a target text unit and a target text are obtained, along with a word-document correspondence table and a candidate sentence library; based on the target text unit, the word-document correspondence table, the target text, and the candidate sentence library, first target advertising information and second target advertising information are obtained, and the target text and / or target text unit, first target advertising information, and / or second target advertising information are provided to the user. Thus, contextual operations enable bidirectional operations of text from complex to simple and from simple to complex, facilitating users to build a global perspective and train their language sense. Furthermore, providing users with advertising information related to the learning text can increase the click-through rate of advertisements and also facilitates users to learn text-related knowledge through advertisements, achieving the goal of quickly mastering the language.

[0011] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit this disclosure. Attached Figure Description

[0012] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this disclosure and, together with the description, serve to explain the principles of this disclosure, and are not intended to unduly limit this disclosure.

[0013] Figure 1 This is a flowchart illustrating a contextual online advertising delivery method according to an exemplary embodiment;

[0014] Figure 2 This is a schematic diagram illustrating the process of context operation matching advertisements in a context-based online advertising delivery method according to an exemplary embodiment;

[0015] Figure 3 This is a flowchart of S3 in a contextual online advertising delivery method according to an exemplary embodiment;

[0016] Figure 4 This is a flowchart of S2 in a contextual online advertising delivery method according to an exemplary embodiment;

[0017] Figure 5 This is a structural diagram of a parsing tree according to an exemplary embodiment;

[0018] Figure 6 This is a flowchart of S2 in another contextual online advertising delivery method according to an exemplary embodiment;

[0019] Figure 7 This is a flowchart of S2 in another contextual online advertising delivery method according to an exemplary embodiment;

[0020] Figure 8This is a flowchart of S4 in a contextual online advertising delivery method according to an exemplary embodiment;

[0021] Figure 9 This is a flowchart of S4 in another contextual online advertising delivery method according to an exemplary embodiment;

[0022] Figure 10 This is a structural diagram illustrating a contextual online advertising delivery device according to an exemplary embodiment;

[0023] Figure 11 This is a structural diagram of a target acquisition unit in a contextual online advertising delivery device according to an exemplary embodiment;

[0024] Figure 12 This is a structural diagram of another target acquisition unit in a contextual online advertising delivery device according to an exemplary embodiment;

[0025] Figure 13 This is a structural diagram of another target acquisition unit in a contextual online advertising delivery device according to an exemplary embodiment;

[0026] Figure 14 This is a structural diagram of a first information acquisition unit in a contextual online advertising delivery device according to an exemplary embodiment;

[0027] Figure 15 This is a structural diagram of a second information acquisition unit in a text context processing apparatus according to an exemplary embodiment;

[0028] Figure 16 This is a structural diagram of a computer system for a server, according to an exemplary embodiment. Detailed Implementation

[0029] To enable those skilled in the art to better understand the technical solutions of this disclosure, the technical solutions in the embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings.

[0030] Unless otherwise required herein, throughout the specification and claims, the term "comprising" is interpreted as open-ended and encompassing, meaning "including, but not limited to." In the description of the specification, terms such as "some embodiments" are intended to indicate that a particular feature, structure, material, or characteristic associated with that embodiment or example is included in at least one embodiment or example of this disclosure. The illustrative representations of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics mentioned may be included in any suitable manner in any one or more embodiments or examples.

[0031] It should be noted that the terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this disclosure are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the present invention described herein can be implemented in orders other than those illustrated or described herein. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this disclosure. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this disclosure as detailed in the appended claims.

[0032] Hereinafter, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, "a plurality of" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0033] It should be noted that the contextual online advertising delivery method of this disclosure can be executed by the contextual online advertising delivery device of this disclosure. This device can be implemented in software and / or hardware, and can be configured in an electronic device. The electronic device can install and run a contextual online advertising delivery program. The electronic device can include, but is not limited to, hardware devices with various operating systems, such as smartphones and tablets.

[0034] Figure 1 This is a flowchart illustrating a contextual online advertising delivery method according to an exemplary embodiment.

[0035] like Figure 1 As shown, the contextual online advertising delivery method provided in this disclosure includes, but is not limited to, the following steps:

[0036] S1: Obtain the text to be processed; the text to be processed includes multiple text units, which are words or phrases.

[0037] It is understood that in this embodiment of the disclosure, the text to be processed can be text for language learning by the user, can be provided by the user, or can be provided by the contextual online advertising delivery device in this embodiment of the disclosure. The text to be processed is used for language learning.

[0038] In the case where the text to be processed is provided by the user, the user can select an article or a paragraph of text and paste it into the corresponding position of the contextual online advertising delivery device in this embodiment of the present disclosure, so that the contextual online advertising delivery device obtains the text to be processed (in the case where the user inputs an article, the article can be pre-processed by segmenting sentences to obtain the text to be processed), and further process the text provided by the user using the contextual online advertising delivery device in this embodiment of the present disclosure.

[0039] In the case where the text to be processed is provided by a contextual online advertising delivery device, in this embodiment of the disclosure, the contextual online advertising delivery device pre-stores text materials for users to learn from. The user selects the corresponding text, and the contextual online advertising delivery device obtains the text selected by the user, obtains the text to be processed, and can further process the text.

[0040] In this embodiment of the disclosure, the text to be processed can be an article or a piece of text. It is understood that the text to be processed includes multiple words or phrases, and may include multiple words, multiple phrases, or at least one word and at least one phrase.

[0041] In this embodiment of the disclosure, in order to facilitate subsequent processing of the text to be processed, there is a constraint on the number of words included in the text to be processed. For example, the number of words included in the text to be processed is limited to between 10 and 35 words, which can reduce the computation time when processing the text to be processed in the future.

[0042] It should be noted that the number of words in the text to be processed can also be limited to other ranges, and can be set according to the hardware environment such as the server. The number of words in the text to be processed can increase with the improvement of hardware performance. This is only for illustration and is not intended to be a specific limitation on the embodiments of this disclosure.

[0043] It should be noted that, in this embodiment of the disclosure, the text to be processed can be English text, or it can also be Chinese text, French text, German text, Italian text, Japanese text, Korean text, etc., and this embodiment of the disclosure does not impose specific limitations on it.

[0044] S2: In response to the user performing context operations on the text to be processed, obtain the target text unit and the target text.

[0045] It should be noted that, in this embodiment of the disclosure, the contextual online advertising delivery device can perform contextual operations on the text to be processed based on the user, and obtain the target text unit and the target text.

[0046] In this embodiment of the disclosure, the contextual online advertising delivery device can process the text to be processed based on the user's contextual operations, and obtain at least one target text unit and at least one target text.

[0047] In this embodiment of the disclosure, the user's contextual operations on the text to be processed may include preceding and following contextual operations, such as... Figure 2 As shown, the above operation can expand the text content of the text to be processed by adding or replacing at least one text unit; the following operation can delete the text content of the text to be processed by deleting at least one text unit.

[0048] In this case, when the user performs a preceding operation on the text to be processed, the newly added or replaced text unit is the target text unit, and the new text generated after the text to be processed is the target text; when the user performs a following operation on the text to be processed, the text unit to be deleted is the target text unit, and the new text generated after the text to be processed is the target text.

[0049] In this embodiment of the disclosure, the user can perform multiple context operations, such as multiple consecutive preceding operations; or multiple consecutive following operations; or multiple preceding and following operations. The order of the preceding and following operations is not required; the user can perform at least one preceding operation followed by at least one following operation, or vice versa. After each preceding or following operation, a target text unit and target text corresponding to that operation can be generated.

[0050] It is understood that in this embodiment of the disclosure, the contextual online advertising delivery device is provided with buttons for users to perform text-below-text operations and text-below-text operations. Correspondingly, the user's text-below-text operation can be performed by clicking the text-below-text operation button, and the user's text-below-text operation can be performed by clicking the text-below-text operation button. Alternatively, the contextual online advertising delivery device is provided with control instructions for users to perform text-below-text operations and text-below-text operations. Correspondingly, the user's text-below-text operation can be performed by triggering the control instruction corresponding to the text-below-text operation, and the user's text-below-text operation can be performed by triggering the control instruction corresponding to the text-below-text operation.

[0051] It should be noted that, in this embodiment of the disclosure, the text to be processed includes multiple text units, which can be words or phrases. Specifically, when the text unit is a word, in this embodiment of the disclosure, the target text unit in the text to be processed is determined based on the part of speech corresponding to the word and the grammatical relationships between different words in the text to be processed.

[0052] For example, taking English text as the text to be processed, when the text unit is a word and the corresponding part of speech is an adjective, the grammatical relationship between the word and other words in the text to be processed is determined. Assuming that the grammatical relationship of the next word adjacent to it is a parent-child relationship such as NP->JJ+NN, which means that the noun phrase is composed of an adjective (JJ) and a noun (NN), then the word with the part of speech of adjective can be determined as the target text unit. The target text unit can be deleted from the text to be processed by the user under the following text operation, and the target text can be generated.

[0053] Therefore, in this embodiment of the disclosure, by using the grammatical relationships between different words or phrases in the text to be processed as constraints, the original grammatical relationships are not destroyed, so that when the text to be processed is subjected to further processing, the result of generating the target text conforms to the grammatical rules.

[0054] S3: Obtain the word-document mapping table and the candidate statement library; wherein, the word-document mapping table includes multiple candidate words and advertising information associated with the candidate words, and the candidate statement library includes multiple candidate statements and advertising information associated with the candidate statements.

[0055] The advertising information may include at least one of the following: the text corresponding to the advertisement, the title, the author, the summary text, and the link.

[0056] The text is the text content of the advertisement; the title is the name of the advertisement; the author is the publisher of the advertisement; the summary text is a summary generated based on the text content of the advertisement, or a part of the text content of the advertisement; and the link is a shortcut link to the advertisement, which allows users to quickly access the specific content of the advertisement by clicking the link.

[0057] like Figure 3 As shown, in some embodiments, the above-described S3 of this disclosure may include the following steps:

[0058] S31: Obtain advertising corpus data; wherein, the advertising corpus data includes at least one of text ads, image ads, and video ads.

[0059] In this embodiment of the disclosure, the advertising corpus data can be crawled from the network using web crawling tools or obtained through self-publishing by advertisers.

[0060] It is understandable that ads scraped or retrieved from the web can be of various types, such as text ads, image ads, video ads, and audio ads.

[0061] In this embodiment of the disclosure, the acquired advertising corpus data includes one or more of text ads, image ads, and video ads.

[0062] S32: Retrieve the ad text and links for text ads, image ads, and video ads.

[0063] In this embodiment of the disclosure, while crawling or obtaining text ads, image ads, and video ads published by advertisers from the network, the corresponding links can be obtained simultaneously, allowing users to quickly access the text ads, image ads, and video ads through the links.

[0064] In this context, obtaining the text of a text ad can be understood as meaning that a text ad includes some text content, and obtaining the text content of the text ad is the way to obtain the text ad's advertising text.

[0065] In this context, obtaining the advertising text of an image advertisement means that the image advertisement includes an image, and the advertising text of the image advertisement is obtained based on the image in the image advertisement. In this embodiment of the disclosure, the advertising text of the image advertisement can be obtained by recognizing the image information in the image advertisement and generating a descriptive text.

[0066] In this context, obtaining the ad text of a video ad can be understood as including images and / or audio in the video ad, and obtaining the ad text of the video ad based on the images and / or audio in the video ad.

[0067] In this embodiment of the disclosure, the advertising text in the video advertisement can be obtained by performing speech recognition on the audio in the video advertisement and obtaining the corresponding text information; the advertising text in the video advertisement can also be obtained by recognizing the image in the video advertisement and generating a descriptive text.

[0068] Of course, in this embodiment of the disclosure, the advertising text of the video advertisement can also be obtained directly from the video advertisement, not limited to the methods of audio recognition and / or image recognition in the above embodiments. The video advertisement can also be directly input into the video description model to directly generate the advertising text, such as template-based video captioning; an attention mechanism is added on the basis of the sequence to sequence model. The attention mechanism distinguishes the importance of certain representations / features by weighting them. For details, please refer to the methods in related technologies, which will not be repeated here.

[0069] It is understandable that video advertisements contain multiple frames of images, and recognizing all of these frames would be computationally intensive. In this embodiment, the images obtained from a video advertisement can be selected from specific frames. Based on this, recognition is performed on the selected specific frames to reduce the computational load.

[0070] In one example, images of specific frames from a video advertisement are retrieved at preset frame intervals, such as 20 frames, 10 frames, etc. Based on this method, images of a subset of frames from the video advertisement are obtained for subsequent recognition, reducing computational load, improving data processing efficiency, and enhancing system performance.

[0071] In another example, an image optical flow-based method identifies at least one specific frame image in a video advertisement. The image optical flow method can be a pyramid-free Lucas-Kanade optical flow method or a pyramid-based Lucas-Kanade optical flow method.

[0072] The method involves extracting feature points from frame images of a video advertisement, calculating optical flow tracking of these feature points, obtaining the optical flow parallax changes of feature points between frames, and assessing the tracking and matching of preceding and subsequent feature points. If the feature point changes in the current frame are small compared to the previous frame, and multiple consecutive frames exhibit similar feature point changes, the current frame is selected as the specific frame image. Based on this method, a subset of frames from the video advertisement can be acquired for subsequent recognition, reducing computational load, improving data processing efficiency, and enhancing system performance.

[0073] In some embodiments, obtaining the advertising text for image ads and video ads includes: obtaining the images in the image ads and video ads; inputting the images into an image description model to generate advertising text.

[0074] In this embodiment of the disclosure, images from image ads and video ads are obtained, and the images are input into a classification model or an image description model to generate ad text.

[0075] The image description model can be the img2txt model. The img2txt model can automatically generate a descriptive text based on an image. The process can be as follows: first, detect the objects in the image (including the category and location of the objects), output the relationships between the objects, and finally express them in reasonable language.

[0076] In some embodiments, inputting an image into an image description model to generate advertising text includes: inputting the image into an encoder, obtaining the category and location of the target object through an object detection model, and generating a feature vector; inputting the feature vector into a decoder to generate advertising text.

[0077] In this embodiment, the image is input to the encoder, and the target object in the image is identified by the detection model to obtain the corresponding vector. The vector is then input to the decoder to map the vector to the corresponding text. Since the vector representation of the image features and the result of the text output are often different in length, it is necessary to map the original sequence to a sequence of different lengths. Based on this, this embodiment uses an encoder-decoder model, which can solve the problem of inconsistent length mapping.

[0078] The Encoder uses a CNN model to extract the corresponding visual features from the image. Then, the Decoder decodes the visual features into an output sequence. The input to the Decoder is word embedding, and the output is the probability of all words in the vocabulary, which in turn generates the corresponding natural language description and advertising text.

[0079] In this embodiment, the basic structure of the CNN model includes two layers: a feature extraction layer, where the input of each neuron is connected to the local receptive field of the previous layer, and the features of that local area are extracted. Once the local features are extracted, their positional relationship with other features is also determined; and a feature mapping layer, where each computational layer of the network consists of multiple feature maps, each of which is a plane with equal weights for all neurons. The sigmoid function is used as the activation function, ensuring the feature maps are shift-invariant. Because neurons share weights, the number of network parameters is reduced. Typical CNN frameworks include LeNet-5, AlexNet, VGG-16, and ResNeXt-50.

[0080] S33: Input the advertising text into the summary generation model to generate summary text.

[0081] Understandably, once the text, image, and video ad texts are obtained, they contain a lot of text content. If the obtained ad texts are directly provided to users, firstly, users may not be able to extract the key points from the large amount of text, resulting in lower user interest in browsing the ad texts; secondly, providing users with a large amount of text content will reduce the number of ad texts available to them, making it inconvenient for users to choose.

[0082] Based on this, in this embodiment of the disclosure, only a portion of the advertising text may be provided to the user. For example, the title, first paragraph, and summary text of the advertising text may be provided to the user.

[0083] In this embodiment, the advertising text is provided to the user as a summary. The advertising text is input into a summary generation model to obtain a summary of the advertising text. The purpose of generating a summary text from the advertising text is to summarize the input advertising text and generate an accurate and concise summary. There are two ways to generate summary text: extraction summarization and abstract summarization. Abstract summarization is not simply copying and pasting text fragments from the input text, but rather generating new words or summarizing important information, thereby making the output summary text fluent and complete.

[0084] For example, a summary generation model could be the transformer model. The transformer model is a deep learning model entirely based on a self-attention mechanism, suitable for parallel computation. It mainly consists of two parts: an encoder and a decoder. When an advertisement text is input, the text data is first encoded by the encoder, and then the encoded data is passed to the decoder module for decoding. After decoding, the translated text is obtained, generating the summary text.

[0085] S34: Generate a word-document mapping table and a candidate sentence library based on the advertising text, summary text, and links.

[0086] In this embodiment of the disclosure, after obtaining the advertising text, summary text, and link, a word-document correspondence table and a candidate statement library are generated based on the advertising text, summary text, and link. The word-document correspondence table includes multiple candidate words and their associated summary text and links, while the candidate statement library includes multiple candidate statements and their associated summary text and links.

[0087] In some embodiments, generating a word-document mapping table based on the advertising text, summary text, and links includes: segmenting the advertising text, removing duplicates, and removing stop words to obtain candidate words; establishing an inverted index of candidate words, summary text, and links to generate a word-document mapping table.

[0088] Understandably, advertising text contains a lot of text content, including multiple sentences. The sentences are segmented to generate a word set, which is then further deduplicated and removed from the word set. Repeated words and stopped words in the word set are removed, and the word set is then filtered to generate multiple candidate words.

[0089] Of course, the above-described method for processing advertising text to obtain candidate words may also include other methods besides those described above. In order to make the obtained candidate words more reasonable, the advertising text may be further processed. This disclosure does not impose specific limitations on this.

[0090] In this embodiment of the disclosure, after obtaining candidate words for the advertising text, an inverted index is established for the candidate words, the summary text, and the links to generate a word-document correspondence table.

[0091] The advertising text is obtained from text ads, image ads, or video ads. In this embodiment, to distinguish different advertising texts, each text is numbered, with each number corresponding one-to-one with the corresponding text ad, image ad, or video ad. In this embodiment, the numbering of text ads, image ads, and video ads is unified as document number. Candidate words are then obtained from the advertising text and numbered, unified as word numbers.

[0092] In this embodiment, an inverted index is established, consisting of candidate words, abstract text, and links. The inverted list records a list of all documents containing a candidate word, along with the word's position within each document. Each record is called a posting. Based on the inverted list, it can be determined which documents contain a particular word.

[0093] For example, if there are a total of 10 text ads, image ads, and video ads, firstly, the ad texts for the text ads, image ads, and video ads are obtained, including 10 ad texts. Each ad text is then numbered to obtain a document number. Next, candidate words in each ad text are obtained. For the convenience of subsequent system processing, each different candidate word needs to be assigned a unique word number, and at the same time, it is recorded which documents contain this word. After such processing, an inverted index is obtained.

[0094] In this embodiment of the disclosure, based on the index of candidate words and advertising text to be ranked, links are added to the corresponding advertising text to generate a word-document correspondence table.

[0095] In some embodiments, generating a candidate statement library based on advertising text, summary text, and links includes: segmenting the advertising text into sentences to obtain candidate statements; establishing a correspondence between candidate statements and summary text and links to generate a candidate statement library.

[0096] Understandably, advertising text contains a lot of text content, including multiple sentences. Preprocessing the advertising text generates multiple candidate sentences.

[0097] The preprocessing of the advertising text includes: deleting the markers in the advertising text, then segmenting it to remove overly long or short sentences, and selecting sentences with a word count within a certain range as candidate sentences.

[0098] For example, sentences with a word count between 10 and 35 are selected as candidate sentences.

[0099] In this embodiment of the disclosure, after obtaining candidate statements of the advertising text, a correspondence between the candidate statements and the summary text and links is established to generate a candidate statement library.

[0100] The advertising text is obtained from text ads, image ads, or video ads. In this embodiment, to distinguish different advertising texts, each text is numbered, and the number corresponds one-to-one with the corresponding text ad, image ad, or video ad. In this embodiment, the numbering of text ads, image ads, and video ads is unified as document number. Candidate statements are then obtained from the advertising text and numbered, unified as statement numbers.

[0101] In this embodiment of the disclosure, a correspondence is established between candidate statements, summary text, and links. Based on the correspondence, a document list containing all documents including a certain candidate statement and the position information of the candidate statement in that document can be obtained.

[0102] For example, if there are a total of 10 text ads, image ads, and video ads, firstly, the ad texts for the text ads, image ads, and video ads are obtained, including 10 ad texts. Each ad text is then numbered, and a document number is obtained. Next, candidate phrases in each ad text are obtained. For the convenience of subsequent system processing, each different candidate phrase needs to be assigned a unique word number, and at the same time, it is recorded which documents contain this candidate phrase. After such processing, the correspondence between candidate phrases and document numbers of ad texts is obtained.

[0103] In this embodiment of the disclosure, based on obtaining the correspondence between the document numbers of candidate statements and advertising texts, links are added to the corresponding advertising texts to generate a candidate statement library.

[0104] S4: Obtain the first target advertising information based on the target text unit and the word document correspondence table, and obtain the second target advertising information based on the target text and the candidate sentence library.

[0105] In this embodiment of the disclosure, after the user performs contextual operations on the text to be processed, based on obtaining the target text unit and the target text, the first target advertising information is obtained according to the target text unit and the word document correspondence table, and the second target advertising information is obtained according to the target text and the candidate sentence library.

[0106] It is understood that the word-document correspondence table includes multiple candidate words and corresponding advertising information. The advertising information may include at least one of the following: the text corresponding to the advertisement, the title, the author, the summary text, and the link. In this embodiment of the disclosure, the user performs contextual operations on the text to be processed, which can obtain one or more target text units, and further obtain the first target advertising information based on the target text units and the word-document correspondence table.

[0107] It is also understood that the candidate statement library includes multiple candidate statements and corresponding advertising information. The advertising information may include at least one of the following: the text corresponding to the advertisement, the title, the author, the summary text, and the link. In this embodiment of the disclosure, the user performs contextual operations on the text to be processed to obtain the target text, and further obtains the second target advertising information based on the target text and the candidate statement library.

[0108] S5: Provide the user with target text and / or target text units, as well as first target advertising information and / or second target advertising information.

[0109] In this embodiment of the disclosure, when the target text and the first target advertising information are obtained, the target text and the first target advertising information are provided to the user; or when the target text and the second target advertising information are obtained, the target text and the second target advertising information are provided to the user; or when the target text, the first target advertising information and the second target advertising information are obtained, the target text, the first target advertising information and the second target advertising information are provided to the user.

[0110] In this embodiment of the disclosure, target text and / or target text units, as well as first target advertising information and / or second target advertising information, can be displayed on the display component of the contextual online advertising delivery device to provide users with target text and / or target text units, as well as first target advertising information and / or second target advertising information.

[0111] It is understood that, in the embodiments of this disclosure, target text units and / or target text may be provided to the user, and may be accompanied by corresponding explanations of the target text and / or target text units.

[0112] For example, taking English text as the text to be processed, the corresponding explanation of the target text unit is provided. For example, if the text to be processed is "blue sky" and the target text unit is the adjective "blue" before the noun "sky", the explanation can be that the part of speech of the target text unit "blue" is an adjective, used to modify the noun "sky".

[0113] Furthermore, it can provide similar words to the target text unit "blue", such as "cerulean", so that users can learn more vocabulary, learn related words such as synonyms or similar words, and improve the user's learning experience.

[0114] It is understood that, in this embodiment of the disclosure, the translation and pronunciation of the target text and target text units may be further provided, allowing users to learn the translated text and pronunciation simultaneously.

[0115] In some embodiments, the contextual online advertising delivery method provided in this disclosure further includes: inputting target text into a syntax analysis model; providing the target text to the user if a matching syntax structure exists; and prompting a contextual operation error and exiting if no matching syntax structure exists.

[0116] In this embodiment of the disclosure, when a user performs contextual operations on the text to be processed to generate target text, the target text needs to be input into a syntax analysis model to determine whether the generated target text can be semantically parsed and whether a matching grammatical structure exists. If a matching grammatical structure exists, the target text is provided to the user; if no matching grammatical structure exists, a contextual operation error is indicated and the process exits. This ensures that the obtained target text conforms to grammatical structure norms and can express a complete meaning, thus avoiding the occurrence of target text that does not conform to language norms and misleads the user's learning.

[0117] Therefore, users can perform contextual operations on the text to be processed. Contextual operations enable two-way operations on the text, from complex to simple and from simple to complex, which helps users build a global view of language structure, train their language sense, deepen their understanding of text structure, and reduce the difficulty of language learning. Furthermore, during the learning process, users can be provided with advertising information related to the learning text. On the one hand, users are more likely to browse the advertisements, which can increase the click-through rate of the advertisements. On the other hand, users can deepen their understanding of the content of the learning text through the advertisements, enrich their related knowledge, and facilitate language learning.

[0118] like Figure 4 As shown, in some embodiments, the above-described S2 of this disclosure may include the following steps:

[0119] S21: In response to the user's operation on the following text of the text to be processed, input the text to be processed into the syntax analysis model and obtain the target syntax structure that matches the text to be processed.

[0120] It is understood that, in this embodiment of the disclosure, before inputting the text to be processed into the syntax analysis model and obtaining the target syntax structure that matches the text to be processed, the method further includes obtaining the syntax analysis model.

[0121] In some embodiments, obtaining a syntax analysis model includes: obtaining corpus text; inputting the corpus text into a grammatical component analysis model based on component analysis to generate a syntax analysis tree; parsing the syntax analysis tree from bottom to top to generate a table structure to obtain a syntax analysis tree library; wherein the syntax analysis tree library includes multiple corpus text units; the table structure includes the relationship between parent and child nodes and the relationship between sibling nodes; assigning weights to the corpus text units to generate a syntax analysis model.

[0122] In this embodiment of the disclosure, the corpus text can be obtained from publicly available articles. The corpus text is obtained by preprocessing the articles. The preprocessing of the articles includes: deleting the markers in the articles, then performing segmentation processing, removing sentences that are too long or too short, and selecting sentences with a word count within a certain range as the corpus text.

[0123] For example, sentences with a word count between 10 and 35 are selected as the corpus text.

[0124] Of course, the acquisition of text corpus in this embodiment is not limited to the above-described examples, and can be set as needed. This embodiment does not impose specific limitations on this.

[0125] In this embodiment of the disclosure, taking English text as an example, the text is input into a grammatical component analysis model based on component analysis. For example, in the case of the text "the medical imaging technology currently has made significant progress in many important domains", a grammatical analysis tree is generated as follows: Figure 5 As shown, leaf nodes are words in a sentence; other non-leaf nodes are the parts of speech of words and phrases composed of words. The characteristic of a parsing tree is that the components closer to the root are the core components of the sentence, while the components closer to the leaf nodes are non-core components.

[0126] In this embodiment of the disclosure, after obtaining the parsing tree corresponding to the corpus text, a parsing tree library is obtained. The parsing tree library includes parsing trees generated from multiple corpus texts, each comprising multiple corpus text units, which are different nodes of the parsing trees. After assigning weights to the corpus text units, a parsing model is generated.

[0127] In one possible implementation, weights are assigned to text units in the corpus in this embodiment. In order to sort the text units in the corpus, a corresponding table is generated by the parsing tree to facilitate the assignment of weights to the text units in the corpus.

[0128]

[0129] Table 1

[0130] In this embodiment of the disclosure, taking English text as an example, the tags used in English grammar parsing adopt the language tag set of the Pennsylvania TreeBank. TreeBank is a large corpus that annotates syntactic and semantic sentence structures, which are usually in the form of trees, hence the name TreeBank.

[0131] The parsing tree representation uses nested parentheses () because it consumes fewer resources and the tree structure is relatively easy to read without software tools. Given a sentence, the grammar can be parsed from left to right. For example, the sentence "the dog run" can be represented as (S(NP(DT the)(NN dog))(VP run)). Its notation is shown in Table 1 above. It should be noted that the above examples are only a partial list and do not include all examples; please refer to the Penn Treebank's linguistic tag set for more details.

[0132]

[0133] Table 2

[0134] For example, the corpus text is "image classification and object detection applications are becoming more robust and more accurate." A syntax analysis tree is generated, and the syntax analysis tree is parsed from bottom to top. Based on the tree structure, a table structure is generated, and the corresponding table structure is shown in Table 2 above.

[0135] It is understood that in this embodiment of the disclosure, there are multiple corpus texts. After performing the above processing on the multiple corpus texts, a corresponding table structure is generated, weights are assigned to the corpus text units, and a syntax analysis model is generated.

[0136] In this embodiment of the disclosure, the symbols and expressions in the syntax analysis model are explained as follows:

[0137] 1) NP->DT+JJ+NN: This means that NP is generated (parsed) into DT, JJ and NN.

[0138] 2) JJ∈(NP->DT+JJ+NN,NP->JJ+NN): means that JJ matches NP->DT+JJ+NN and NP->JJ+NN.

[0139] 3) JJ∈(NP->DT+JJ+NN,NP->JJ+NN)&(ORDER(1))): indicates that JJ matches NP->DT+JJ+NN, NP->JJ+NN, and the weight level is 1.

[0140] 4) JJ∈((NP->DT+JJ+NN,NP->JJ+NN)&(ORDER(1))) / JJ: This means that JJ matches NP->DT+JJ+NN, NP->JJ+NN, and the percentage of all JJ records with a weight level of 1.

[0141] Based on the above symbol and expression conventions, the following matching structures account for a certain percentage:

[0142] JJ∈((NP->DT+JJ+NN,NP->JJ+NN)&(ORDER(1))) / JJ;

[0143] RB∈((ADVP->RB)&(ORDER(2))) / RB;

[0144] PP∈((VP->VBN+NP+PP)&(ORDER(3))) / PP;….

[0145] Based on the above structure, statistical analysis is performed on the database that generates the corresponding tables to generate a syntax analysis model:

[0146] Model={JJ∈(NP->DT+JJ+NN, NP->JJ+NN)&(ORDER(1))) / JJ,...,}.

[0147] Therefore, based on the obtained syntax analysis model, the embodiments of this disclosure can obtain the target syntax structure that matches the text to be processed by inputting the text to be processed into the syntax analysis model.

[0148] S22: Based on the target grammatical structure, obtain the weight levels corresponding to multiple text units in the text to be processed.

[0149] S23: When the weighting levels include at least two levels, determine the first target text unit among multiple text units.

[0150] In this embodiment of the disclosure, after inputting the text to be processed into the syntax analysis model and obtaining the target syntax structure matching the text, it is possible to obtain multiple text units corresponding to the text and the weight levels corresponding to the text units. Therefore, based on the determined weight levels of the text units, the first target text unit among the multiple text units can be determined.

[0151] Understandably, if the weight level of the text to be processed is only one level, and the user attempts to add text, since there is only one level, it means that the text to be processed is already the most basic text structure and cannot be added. In this case, even if the user attempts to add text, the target text unit cannot be obtained.

[0152] When multiple weight levels are obtained for the text to be processed, the user can determine the lowest-weighted text unit as the first target text unit for each subsequent operation on the text to be processed.

[0153] S24: Delete the first target text unit in the text to be processed to generate the target simplified text.

[0154] S25: Obtain the first target text unit as the target text unit, and obtain the target simplified text as the target text.

[0155] In this embodiment of the disclosure, when a first target text unit is determined, the first target text unit in the text to be processed is deleted to generate a target simplified text.

[0156] It should be noted that when there are multiple weight levels for the text to be processed, for example, when there are 3 weight levels, the user can perform multiple text-to-text operations. In the first text-to-text operation, the first target text unit is determined to be a text unit of the third level. In the second text-to-text operation, the first target text unit is determined to be a text unit of the second level. At this time, if the user continues to perform text-to-text operations, the first target text unit will not be able to be obtained.

[0157] Based on this, each time the user performs a text operation, the first target text unit is deleted from the text to be processed, generating the target simplified text. This process continues until the first target text unit is obtained. In the event that the text units in the text to be processed constitute the target simplified text with the most basic structure.

[0158] It is understood that, in this embodiment of the disclosure, the text to be processed is English text, and the target simplified text with the most basic structure is the five basic sentence patterns of English.

[0159] For example, the five basic sentence structures are as follows:

[0160] S+V subject-verb structure; in this sentence structure, V is an intransitive verb, also called an intransitive verb (vi).

[0161] The sentence structure is S+V+F (subject-linking verb-complement). In this sentence structure, V is a linking verb. Common linking verbs include: look, seem, appear, sound, feel, taste, smell, grow, get, fall, fall / asleep, stand / sit still, become, turn, etc.

[0162] The sentence structure is S+V+O (subject-verb-object). In this sentence structure, V is a transitive verb (vt.), so it has an object.

[0163] The sentence structure is S+V+O1+O2, where V is a transitive verb with two objects. Common verbs that require two objects include give, ask, bring, offer, send, pay, lend, show, tell, buy, get, rob, warn, etc.

[0164] S+V+O+C Subject-Verb-Object-Complement Structure.

[0165] Where S = subject; V = predicate; P = complement; O = object; O1 = indirect object; O2 = direct object; C = object complement.

[0166] Of course, similar concepts can be used for other texts besides English text, and this disclosure does not impose specific limitations on them.

[0167] In some embodiments, if there is only one weight level, it is determined that there is no first target text unit, and the text to be processed is indicated as the most basic structure text and the process is exited.

[0168] Understandably, when the text to be processed is input into the syntax analysis model, the target syntax structure that matches the text to be processed is obtained. Based on the target syntax structure, the weight levels corresponding to the multiple text units included in the text to be processed are determined. If there is only one weight level, it is determined that there is no first target text unit. In this case, the text to be processed is indicated as the most basic structure text and the process exits.

[0169] Taking English text as an example, if the text to be processed is in the five basic sentence structures of English, it is determined that there is no first target text unit in the text to be processed. At this time, the text to be processed is prompted as the most basic structure text and the process is exited.

[0170] like Figure 6 As shown, in some embodiments, the above-described S2 of this disclosure may include the following steps:

[0171] S201: Obtain the restricted vocabulary list; wherein the restricted vocabulary list includes multiple restricted words.

[0172] S202: In response to the user's first contextual operation on the text to be processed, input the text to be processed into the text generation model, obtain at least one editing operation based on the restricted vocabulary, and generate the target generated text.

[0173] S203: Identify the text units that are different between the target generated text and the text to be processed as the second target text units.

[0174] In response to the user's previous operation on the text to be processed, the text to be processed is input into the text generation model to generate a second target text unit and a target generated text.

[0175] S204: Obtain the second target text unit as the target text unit, and obtain the target generated text as the target text.

[0176] The text generation model can be a Lasertagger model. The text to be processed is input into the Lasertagger model, which generates a series of editing operations to replace the text units in the original text, producing text that better suits the application scenario. The four editing operations used are: Keep (copying a text unit to the output), Delete (deleting a text unit), Add (adding a text unit), and Swap (changing the order of two text units).

[0177] The added text units all come from a restricted vocabulary. By restricting the vocabulary, the size of the vocabulary can be minimized and the number of training samples can be maximized, including only the necessary text units that need to be added to the text to be processed; the text units can be words or phrases.

[0178] Limiting the number of text units in the vocabulary reduces the decision load on the corresponding output and prevents the model from arbitrarily adding text units. Since the input and output texts highly overlap, only a subset of text units need to be modified. This allows for accurate parallel prediction of editing operations, significantly improving the speed of end-to-end text generation.

[0179] In this embodiment of the disclosure, a restricted vocabulary can be constructed based on a specific domain or direction, so that when a user performs a first context operation on the text to be processed, a second target text unit can be obtained in a targeted manner, and the target generated text can be obtained. Furthermore, when obtaining the first target advertising information based on the second target text unit and obtaining the second target advertising information based on the target generated text, the obtained advertising information can be associated with a specific domain or direction, providing the user with relevant advertising information, and can be better applied to advertising promotion.

[0180] In some embodiments, a first number of text units included in the target generated text is obtained; if the first number is greater than a first preset threshold, in response to the user's first context operation on the text to be processed, a prompt is made indicating that the generated text has reached the upper limit and the process exits.

[0181] It is understood that in this embodiment of the disclosure, the user can perform the first above operation multiple times. As the target generated text is generated multiple times, the number of text units included in the obtained target generated text will also increase. The more text units included, the more time the system needs to process the data, and the lower the computing efficiency will be.

[0182] Based on this, in this embodiment of the disclosure, the text unit data included in the target generated text are statistically analyzed to obtain the first number of text units included in the target generated text. If the first number is greater than the first preset threshold, and if there is a user's first above-mentioned operation, the generated text is prompted to reach the upper limit and the user exits.

[0183] The first preset threshold can be 100, 80, or 50, etc., and can be set according to the server computing power and network bandwidth used by the system. This embodiment does not impose specific restrictions on this.

[0184] like Figure 7 As shown, in some embodiments, the above-described S2 of this disclosure may include the following steps:

[0185] S2001: In response to the user's second context operation on the text to be processed, the text to be processed is divided into at least one text unit sequence according to preset conditions.

[0186] In this embodiment of the disclosure, based on the text to be processed, in response to a second context operation by the user on the text to be processed, text units can be added to the text to be processed to enrich the text to be processed, facilitate user learning, and improve the user experience.

[0187] The preset condition can be to divide a preset number of adjacent text units into a text unit sequence. The text unit can be a word or a phrase.

[0188] For example, the preset number can be two, four, or six, etc., and this disclosure does not impose specific limitations on this.

[0189] In one possible implementation, taking English text as an example, for instance, the text to be processed is: "size are important for a model".

[0190] Starting from the beginning of the sentence, a sliding window is used to predict every 2 or 4 words as a text unit sequence. This example uses 4 words. (This parameter can be set according to the system's training model).

[0191] The sliding window illustration shows that the first text unit sequence generated by the first sliding window is "size are important for", the second text unit sequence generated by the second sliding window is "are important for a", and the third text unit sequence generated by the third sliding window is "important for a model".

[0192] S2002: Input the text unit sequence into the trained word vector model to predict the third target text unit.

[0193] In this embodiment of the disclosure, the trained word vector model can be a trained DistributedRepresentation encoding model. The text unit sequence is sequentially input into the trained word vector model to predict the corresponding third target text unit, as exemplified in Table 3 below:

[0194] previous2 previous1 next1 next2 out The second one before The first one before The first one after the next The second one after that Output Sample size are important none size are important for very are important for a none import for a model none for a model learning

[0195] Table 3

[0196] The third target text unit is the text unit in the "out / output" column of Table 3 above.

[0197] S2003: Traverse the text to be processed and determine the first reserved position in the text to be processed corresponding to the third target text unit.

[0198] In this embodiment of the disclosure, the text to be processed is traversed to obtain the first reserved position in the text to be processed corresponding to the third target text unit.

[0199] S2004: Add the third target text unit to the first reserved position of the text to be processed, and generate the target new text.

[0200] S2005: Get the third target text unit as the target text unit, and get the target newly added text as the target text.

[0201] In this embodiment of the disclosure, in response to the user's first second preceding text operation, the target new text is generated: "Samplesize are very important for a learning model". Here, "very" and "learning" are the third target text units.

[0202] Based on the first second pre-text operation, in response to the user's second second pre-text operation, the above process continues to obtain the text unit sequence, and then inputs the text unit sequence into the trained word vector model to predict the third target text unit, as shown in Table 4 below:

[0203] previous2 previous1 next1 next2 out The second one before The first one before The first one after the next The second one after that Output Sample size are very none … … … … none for a learning a machine a learning model none

[0204] Table 4

[0205] In response to the user's second operation above the second text, the target new text is generated: "sample size are very important for a machine learning model", with "machine" being the third target text unit, generated based on the previous one.

[0206] It should be noted that the above examples are for illustrative purposes only. In this embodiment of the disclosure, the user may perform the second above operation multiple times, and this embodiment of the disclosure does not impose any specific restrictions on this.

[0207] Based on this, in this embodiment of the disclosure, without changing the original syntax and semantics, in response to the user's second context operation on the text to be processed, additional text is added to the text to be processed, which can enrich the user's vocabulary.

[0208] In some embodiments, a second number of text units included in the target newly added text is obtained; if the second number is greater than a second preset threshold, in response to the user's second preceding operation, a prompt is made that the newly added text has reached the limit and the user exits.

[0209] It is understood that in this embodiment of the present disclosure, the user can perform multiple second text operations. With each second text operation, the number of text units included in the newly acquired target text will increase. The more text units included, the more time the system needs to process the data, and the lower the computational efficiency will be.

[0210] Based on this, in this embodiment of the disclosure, the text unit data included in the target newly added text are statistically analyzed to obtain the second number of text units included in the target newly added text. If the second number is greater than the second preset threshold, and there is a second previous operation by the user, the system prompts that the newly added text has reached the limit and exits.

[0211] The second preset threshold can be 100, 80, or 50, etc., and can be set according to the server computing power and network bandwidth used by the system. This embodiment does not impose specific restrictions on this.

[0212] In some embodiments, the second preset threshold is equal to the first preset threshold.

[0213] like Figure 8 As shown, in some embodiments, the above-described S4 of this disclosure may include the following steps:

[0214] S41: Input the target text unit into the trained word vector model to generate the target word vector.

[0215] In this embodiment of the disclosure, the trained word vector model can be a trained DistributedRepresentation encoding model. By inputting the target text unit into the trained word vector model, the target word vector can be generated.

[0216] S42: Calculate the similarity between the target word vector and the candidate word vectors generated from the candidate words in the word-document correspondence table.

[0217] In this embodiment of the disclosure, candidate words are input into a trained word vector model to generate candidate word vectors.

[0218] There are multiple candidate words, which can yield multiple candidate word vectors. The similarity between the target word vector of the target text unit and each candidate word vector is calculated sequentially.

[0219] S43: Based on similarity, determine the advertising information associated with the candidate word that has the highest similarity to the target word vector as the first target advertising information.

[0220] In this embodiment of the disclosure, when the similarity between the target word vector of the target text unit and each candidate word vector is obtained, a candidate word vector with the highest similarity can be determined based on the similarity, and the advertising information associated with the candidate word vector with the highest similarity can be determined based on the word-document correspondence table. Then, the advertising information associated with the candidate word vector with the highest similarity is determined as the first target advertising information.

[0221] In some embodiments, when multiple target text units are obtained, obtaining first target advertising information based on the target text units and a word-document lookup table includes: traversing multiple target text units, obtaining target text units whose word attributes are nouns or adjectives, matching them with candidate words in the word-document lookup table, and obtaining first target advertising information.

[0222] Understandably, advertisements generally focus on things, providing relevant information about a particular thing. As a result, the most common content in advertisements consists of words with noun or adjective attributes. By filtering multiple target text units for words with noun or adjective attributes, targeted advertising information can be prioritized, avoiding cluttered primary target advertising information and improving the click-through rate of advertisements.

[0223] In some embodiments, the contextual online advertising delivery method provided in this disclosure further includes: obtaining a trained word vector model, wherein the method includes: obtaining a training dataset; inputting the training dataset into the word vector model to train the word vector model and generate a trained word vector model.

[0224] In this embodiment of the disclosure, taking English text as an example, a training dataset is obtained. The corpus can be a publicly available English novel in text format. The sentences in the English novel are segmented into words. Based on the word segmentation, a training dataset is generated through a sliding window (the window length can be set; for example, it is set to the target word and the two adjacent words before and after it, for a total of four adjacent words).

[0225] It should be noted that the window length can be set. The window length can be 3 to retrieve the target word and one word immediately before and after it, for a total of two adjacent words, generating the training dataset. Alternatively, the window length can be 7, etc., depending on the requirements.

[0226] In one possible implementation, the training dataset is obtained, and then input into the word vector model. The method for training the word vector model is as follows:

[0227] Example: The technology currently has made significant progress in many important domains. The entire sentence is traversed using a sliding window. For example, if the sliding window length is set to 5, then the first two and last two words of each word serve as input, and the output is the target word.

[0228] The training dataset generated from the example sentences is shown in Table 5 below:

[0229] previous2 previous1 next1 next2 out The second one before The first one before The first one after the next The second one after that Output / / technology currently the / the currently has technology the technology has made currently technology currently made significant has currently has significant progress made has made progress in significant made significant in many progress significant progress many important in progress in important domains many in many domains / important many important / / domains

[0230] Table 5

[0231] In this embodiment of the disclosure, the word vector model can be a Distributed Representation encoding model. After obtaining the training dataset, the training dataset is sequentially input into the Distributed Representation encoding model to train the word vector model and generate a trained word vector model.

[0232] like Figure 9 As shown, in some embodiments, the above-described S4 of this disclosure may further include the following steps:

[0233] S401: Input the target text into the statement vector model to generate the target text vector.

[0234] The sentence vector model can be a doc2vec model, a Bag of Words model, a TF-IDF model, a BERT model, etc. Bag of Words (BOW): This constructs text vectors based on the frequency of each word in the text; the vector size is equal to the vocabulary size. A tool that can be used is doc2bow in gensim. TF-IDF: Building upon BOW, this considers the importance of each word, while maintaining the same vector size as the vocabulary. A tool that can be used is TFIDFModel in gensim. In this embodiment, inputting the target text into the sentence vector model allows the acquisition of the target text vector.

[0235] S402: Calculate the similarity between the target text vector and the candidate statement vector generated from the candidate statements in the candidate statement library.

[0236] In this embodiment of the disclosure, candidate statements are input into a statement vector model to obtain candidate statement vectors corresponding to the candidate statements. Since there are multiple candidate statements, multiple candidate statement vectors can be obtained. The similarity between the target text vector of the target text unit and each candidate statement vector is calculated sequentially.

[0237] S403: Based on similarity, determine the advertising information associated with the candidate statement that has the highest similarity to the target text vector as the second target advertising information.

[0238] In this embodiment of the disclosure, when the similarity between the target text vector of the target text and each candidate sentence vector is obtained, a candidate sentence vector with the highest similarity can be determined based on the similarity, and the advertising information associated with the candidate sentence vector with the highest similarity can be determined based on the candidate sentence library. Then, the advertising information associated with the candidate sentence vector with the highest similarity is determined as the second target advertising information.

[0239] In some embodiments, when there are multiple first target advertisements and / or multiple second target advertisements, providing the first target advertisements and / or second target advertisements to the user includes: sorting the multiple first target advertisements and / or multiple second target advertisements according to a preset rule; and selecting to provide the user with a preset number of first target advertisements and / or second target advertisements that are ranked first.

[0240] In this embodiment of the disclosure, when the target text and / or target text units, as well as the first target advertising information and / or the second target advertising information, are displayed by the display component of the contextual online advertising delivery device to provide the user with the target text and / or target text units, as well as the first target advertising information and / or the second target advertising information, it is conceivable that the content that the display component of the contextual online advertising delivery device can display is limited.

[0241] Based on this, when multiple first-target advertising information and / or multiple second-target advertising information are obtained, the multiple first-target advertising information and / or multiple second-target advertising information can be sorted according to preset rules; and the user can be selected to be provided with a preset number of first-target advertising information and / or second-target advertising information that are ranked first.

[0242] The preset rules can include ad clicks, ad views, random sorting, and ad information category sorting. The preset number can be determined based on the content size that the display component of the online advertising device can display, and this embodiment does not impose specific limitations on this.

[0243] In this embodiment of the disclosure, a preset number of first target advertisements and / or second target advertisements are selected to be provided to the user, and a recall method is adopted, such as collaborative filtering, FM (Factorization Machine), FFM (Field-aware Factorization Machines), graph model, dual-tower model, DNN model, Deep Retrieval algorithm, etc.

[0244] Understandably, content ranked higher in the list is more relevant to the text being processed and represents content of interest to the user. In this embodiment, a ranking model using deep learning algorithms can also be used. By optimizing the model and tuning its parameters, or by inputting a large amount of valuable sample data, a relatively stable level can be achieved through training.

[0245] The retrieved results are sorted, and the top k (k is usually a single digit) results are used as the final output of the ad recommendation system. Commonly used algorithms in the sorting stage include Logistic Regression (LR), Factorization Machine (FM), and DeepFM. During user interactions, the user can browse returned ad information, summary text, links, etc., based on the recall and sorting results from the recommendation system.

[0246] Figure 10 This is a structural diagram of a contextual online advertising delivery device according to an exemplary embodiment.

[0247] like Figure 10 As shown, the contextual online advertising delivery device 1 includes: a text acquisition unit 11, a target acquisition unit 12, a data acquisition unit 13, a first information acquisition unit 14, a second information acquisition unit 15, and an information provision unit 16.

[0248] The text acquisition unit 11 is used to acquire the text to be processed; wherein the text to be processed includes multiple text units, and the text units are words or phrases.

[0249] The target acquisition unit 12 is used to acquire the target text unit and the target text in response to the user performing a context operation on the text to be processed.

[0250] The data acquisition unit 13 is used to acquire a word-document correspondence table and a candidate statement library; wherein, the word-document correspondence table includes multiple candidate words and advertising information associated with the candidate words, and the candidate statement library includes multiple candidate statements and advertising information associated with the candidate statements.

[0251] The first information acquisition unit 14 is used to acquire the first target advertising information based on the target text unit and the word document correspondence table.

[0252] The second information acquisition unit 15 is used to acquire second target advertising information based on the target text and the candidate statement library.

[0253] And information providing unit 16, for providing users with target text and / or target text units, as well as first target advertising information and / or second target advertising information.

[0254] like Figure 11 As shown, in some embodiments, the target acquisition unit 12 includes: a syntax structure acquisition module 121, a weight level acquisition module 122, a first target text unit determination module 123, a target simplified text generation module 124, and a first data acquisition module 125.

[0255] The syntax structure acquisition module 121 is used to respond to the user's following operation on the text to be processed, input the text to be processed into the syntax analysis model, and obtain the target syntax structure that matches the text to be processed.

[0256] The weight level acquisition module 122 is used to obtain the weight level corresponding to multiple text units in the text to be processed according to the target grammatical structure.

[0257] The first target text unit determination module 123 is used to determine the first target text unit among a plurality of text units when the weight level includes at least two levels.

[0258] The target simplified text generation module 124 is used to delete the first target text unit in the text to be processed and generate the target simplified text.

[0259] The first data acquisition module 125 is used to acquire the first target text unit as the target text unit and to acquire the target simplified text as the target text.

[0260] like Figure 12As shown, in some embodiments, the target acquisition unit 12 includes: a restricted vocabulary acquisition module 126, a second target data generation module 127, and a first data acquisition module 128.

[0261] The restricted vocabulary acquisition module 126 is used to acquire a restricted vocabulary; wherein, the restricted vocabulary includes multiple restricted words.

[0262] The second target data generation module 127 is used to respond to the user's first context operation on the text to be processed, input the text to be processed into the text generation model, obtain at least one editing operation according to the restricted vocabulary, generate target generated text, and obtain at least one second target text unit that is different from the text to be processed.

[0263] The second data acquisition module 128 is used to acquire the second target text unit as the target text unit and to acquire the target generated text as the target text.

[0264] like Figure 13 As shown, in some embodiments, the target acquisition unit 12 includes: a text unit sequence acquisition module 1201, a third target text unit generation module 1202, a location acquisition module 1203, a target new text generation module 1204, and a third data acquisition module 1205.

[0265] The text unit sequence acquisition module 1201 is used to divide the text to be processed into at least one text unit sequence according to preset conditions in response to a second context operation of the user on the text to be processed.

[0266] The third target text unit generation module 1202 is used to input the text unit sequence into the trained word vector model to predict the third target text unit.

[0267] The location acquisition module 1203 is used to traverse the text to be processed and determine the first reserved position in the text to be processed corresponding to the third target text unit.

[0268] The target new text generation module 1204 is used to add the third target text unit to the first reserved position of the text to be processed, and generate the target new text.

[0269] The third data acquisition module 1205 is used to acquire the third target text unit as the target text unit and to acquire the target newly added text as the target text.

[0270] In some embodiments, the data acquisition unit 13 is specifically used to acquire advertising corpus data; wherein the advertising corpus data includes at least one of text ads, image ads, and video ads; acquire the advertising text and links of text ads, image ads, and video ads; input the advertising text into the summary generation model to generate summary text; and generate a word-document correspondence table and a candidate sentence library based on the advertising text, summary text, and links.

[0271] In some embodiments, the data acquisition unit 13 is further configured to acquire images from image advertisements and video advertisements; input the images into an image description model to generate advertisement text.

[0272] In some embodiments, the data acquisition unit 13 is further configured to input an image to an encoder, obtain the category and location of the target object through a classification model or an object detection model, generate a feature vector, and input the feature vector to a decoder to generate advertising text.

[0273] In some embodiments, the data acquisition unit 13 is further configured to perform word segmentation, deduplication, and stop word removal on the advertising text to obtain candidate words; establish an inverted index of candidate words, summary text, and links to generate a word-document correspondence table.

[0274] In some embodiments, the data acquisition unit 13 is further configured to segment the advertising text into sentences, obtain candidate sentences, establish a correspondence between the candidate sentences and the summary text and links, and generate a candidate sentence library.

[0275] like Figure 14 As shown, in some embodiments, the first information acquisition unit 14 includes: a target word vector generation module 141, a first similarity calculation module 142, and a first target advertising information determination module 143.

[0276] The target word vector generation module 141 is used to input the target text unit into the trained word vector model to generate the target word vector.

[0277] The first similarity calculation module 142 is used to calculate the similarity between the target word vector and the candidate word vector generated from the candidate words in the word-document correspondence table.

[0278] The first target advertising information determination module 143 is used to determine the advertising information associated with the candidate word with the highest similarity to the target word vector as the first target advertising information based on similarity.

[0279] like Figure 15 As shown, in some embodiments, the second information acquisition unit 15 includes: a target text vector generation module 151, a second similarity calculation module 152, and a second target advertising information determination module 153.

[0280] The target text vector generation module 151 is used to input the target text into the statement vector model and generate the target text vector.

[0281] The second similarity calculation module 152 is used to calculate the similarity between the target text vector and the candidate sentence vector generated from the candidate sentences in the candidate sentence library.

[0282] The second target advertising information determination module 153 is used to determine the advertising information associated with the candidate statement with the highest similarity to the target text vector as the second target advertising information based on similarity.

[0283] In some embodiments, when multiple target text units are acquired, the first information acquisition unit 14 is specifically used to traverse multiple target text units, acquire target text units whose word attributes are nouns or adjectives, match them with candidate words in a word document lookup table, and acquire first target advertising information.

[0284] In some embodiments, when there are multiple first target advertisements and / or multiple second target advertisements, the information providing unit 16 is specifically used to sort the multiple first target advertisements and / or multiple second target advertisements according to a preset rule; and select to provide the user with a preset number of first target advertisements and / or second target advertisements that are ranked first.

[0285] In some embodiments, the information providing unit 16 is further configured to input the target text into the syntax analysis model, and provide the target text to the user if a matching syntax structure exists; and prompt a context operation error and exit if no matching syntax structure exists.

[0286] Regarding the apparatus in the above embodiments, the specific manner in which each module performs its operation has been described in detail in the embodiments related to the method, and will not be elaborated upon here.

[0287] The beneficial effects that the contextual online advertising delivery device provided in this embodiment can achieve are the same as those that the contextual online advertising delivery method provided in the above examples can achieve, and will not be repeated here.

[0288] Figure 16 This is a structural diagram of a computer system 600 for a contextual online advertising delivery method, according to an exemplary embodiment.

[0289] Figure 16 The server shown is merely an example and should not be construed as limiting the functionality and scope of use of the embodiments disclosed herein.

[0290] Computer system 600 includes a central processing unit (CPU) 601, which performs various appropriate actions and processes based on programs stored in read-only memory (ROM) 602 or programs loaded from storage section 606 into random access memory (RAM) 603. RAM 603 also stores various programs and data required for the operation of system 600. CPU 601, ROM 602, and RAM 603 are interconnected via bus 604. Input / output (I / O) interface 605 is also connected to bus 604.

[0291] The following components are connected to I / O interface 605: storage section 606, including hard disks, etc.; and communication section 607, including network interface cards such as LAN (Local Area Network) cards, modems, etc. Communication section 607 performs communication processing via a network such as the Internet. Drive 608 is also connected to I / O interface 605 as needed. Removable media 609, such as disks, optical disks, magneto-optical disks, semiconductor memories, etc., are installed on drive 608 as needed so that computer programs read from them can be installed into storage section 606 as needed.

[0292] In particular, according to embodiments of this disclosure, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of this disclosure include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication section 607, and / or installed from removable medium 609. When the computer program is executed by central processing unit (CPU) 601, it performs the functions defined above in the methods of this disclosure.

[0293] It should be noted that the computer-readable medium described in this disclosure can be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. A computer-readable storage medium can be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this disclosure, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In this disclosure, a computer-readable signal medium can include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. Computer-readable signal media can also be any computer-readable medium other than computer-readable storage media, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to: wireless, wire, optical fiber, RF, etc., or any suitable combination thereof.

[0294] Computer program code for performing the operations of this disclosure can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0295] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0296] The units described in the embodiments of this disclosure can be implemented in software or hardware. The described units can also be located in a processor; for example, a processor can be described as including a receiving unit, an acquisition unit, an establishment unit, and a matching unit. The names of these units do not necessarily limit the specific unit; for example, a receiving unit can also be described as a "unit for receiving statistical requests."

[0297] To implement the above embodiments, this disclosure also proposes a storage medium.

[0298] When the instructions in the storage medium are executed by the processor of the electronic device, the electronic device is able to perform the context-based online advertising delivery method described above. For example, the storage medium may be ROM (Read Only Memory Image), RAM (Random Access Memory), CD-ROM (Compact Disc Read-Only Memory), magnetic tape, floppy disk, and optical data storage device, etc.

[0299] To implement the above embodiments, this disclosure also provides a computer program product that, when executed by the processor of an electronic device, enables the electronic device to perform the contextual online advertising delivery method as described above.

[0300] Other embodiments of this disclosure will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This disclosure is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of this disclosure are indicated by the following claims.

[0301] It should be understood that this disclosure is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this disclosure is limited only by the appended claims.

Claims

1. A contextual online advertising delivery method, characterized in that, The method is applied to language learning applications, including: Obtain the text to be processed; wherein, the text to be processed is English text for language learning by the user, and the text to be processed includes multiple text units, wherein the text unit is a word; In response to a user performing a context operation on the text to be processed, the target text unit and the target text are obtained. The user's context operation on the text to be processed includes a preceding operation and a following operation. The preceding operation expands the text content of the text to be processed by adding or replacing at least one text unit in the text to be processed. The following operation deletes the text content of the text to be processed by deleting at least one text unit in the text to be processed. Obtain a word-document mapping table and a candidate statement library; wherein, the word-document mapping table includes multiple candidate words and advertising information associated with the candidate words, and the candidate statement library includes multiple candidate statements and advertising information associated with the candidate statements; Based on the target text unit and the word-document correspondence table, obtain the first target advertising information; Based on the target text and the candidate statement library, obtain the second target advertisement information; and Provide the user with the target text and / or the target text unit, as well as the first target advertising information and / or the second target advertising information; The step of responding to user input of text to be processed, performing context operations on the text to be processed, and obtaining target text units and target text includes: In response to the user's subsequent operation on the text to be processed, the text to be processed is input into the syntax analysis model to obtain the target syntax structure that matches the text to be processed; Based on the target grammatical structure, obtain the weight levels corresponding to multiple text units in the text to be processed; When the weighting levels include at least two levels, a first target text unit is determined among the plurality of text units; The first target text unit in the text to be processed is deleted to generate the target simplified text; The first target text unit is used as the target text unit, and the target simplified text is used as the target text; The method further includes: The target text is input into the syntax analysis model, and if a matching syntax structure exists, the target text is provided to the user. If no matching grammatical structure exists, a context operation error will be displayed and the program will exit.

2. The method according to claim 1, characterized in that, The step of responding to user input of text to be processed, performing context operations on the text to be processed, and obtaining target text units and target text includes: Obtain a restricted vocabulary list; wherein the restricted vocabulary list includes multiple restricted words; In response to the user's first contextual operation on the text to be processed, the text to be processed is input into the text generation model, and at least one editing operation is obtained according to the restricted vocabulary to generate the target generated text; The text unit that is different from the target generated text and the text to be processed is identified as the second target text unit; The second target text unit is used as the target text unit, and the target generated text is used as the target text.

3. The method according to claim 1, characterized in that, The step of responding to user input of text to be processed, performing context operations on the text to be processed, and obtaining target text units and target text includes: In response to the user’s second contextual operation on the text to be processed, the text to be processed is divided into at least one text unit sequence according to preset conditions; The text unit sequence is input into a trained word vector model to predict the third target text unit. Traverse the text to be processed and determine the first reserved position in the text to be processed corresponding to the third target text unit; The third target text unit is added to the first reserved position of the text to be processed to generate the new target text; The third target text unit is used as the target text unit, and the newly added target text is used as the target text.

4. The method according to claim 1, characterized in that, The process of obtaining the word-document correspondence table and the candidate sentence library includes: Acquire advertising corpus data; wherein the advertising corpus data includes at least one of text ads, image ads, and video ads; Obtain the advertising text and links of the text ads, image ads, and video ads; The advertising text is input into the summary generation model to generate summary text; Based on the advertising text, the summary text, and the link, generate the word-document correspondence table and the candidate sentence library.

5. The method according to claim 4, characterized in that, The step of obtaining the advertising text of the image advertisement and the video advertisement includes: Obtain the images from the image advertisement and the video advertisement; The image is input into the image description model to generate the advertising text.

6. The method according to claim 5, characterized in that, The step of inputting the image into an image description model to generate the advertising text includes: The image is input into the encoder, and the category and location of the target object are obtained through a classification model or an object detection model to generate a feature vector; The feature vector is input into the decoder to generate the advertising text.

7. The method according to any one of claims 4 to 6, characterized in that, The step of generating the word-document mapping table based on the advertisement text, the summary text, and the link includes: The advertising text is segmented, deduplicated, and stop words are removed to obtain the candidate words; An inverted index is created to correspond the candidate words with the summary text and the links, generating a word-document mapping table.

8. The method according to any one of claims 4 to 6, characterized in that, The step of generating the candidate statement library based on the advertisement text, the summary text, and the link includes: The advertisement text is segmented into sentences to obtain the candidate sentences; Establish the correspondence between the candidate statements, the summary text, and the links to generate the candidate statement library.

9. The method according to claim 1, characterized in that, The step of obtaining the first target advertising information based on the target text unit and the word-document correspondence table includes: The target text unit is input into the trained word vector model to generate the target word vector; Calculate the similarity between the target word vector and the candidate word vector generated from the candidate words in the word-document correspondence table; Based on the similarity, the advertising information associated with the candidate word that has the highest similarity to the target word vector is determined as the first target advertising information.

10. The method according to claim 1, characterized in that, The step of obtaining the second target advertisement information based on the target text and the candidate statement library includes: The target text is input into the sentence vector model to generate a target text vector. Calculate the similarity between the target text vector and the candidate sentence vector generated from the candidate sentences in the candidate sentence library; Based on the similarity, the advertising information associated with the candidate statement that has the highest similarity to the target text vector is determined as the second target advertising information.

11. The method according to claim 1, characterized in that, In the case of obtaining multiple target text units, obtaining the first target advertising information based on the target text units and the word-document lookup table includes: Traverse multiple target text units, obtain target text units with word attributes of noun or adjective, match them with candidate words in the word document lookup table, and obtain the first target advertising information.

12. The method according to claim 1, characterized in that, In the presence of multiple first target advertising messages and / or multiple second target advertising messages, providing the user with the first target advertising message and / or the second target advertising message includes: The multiple first target advertisements and / or multiple second target advertisements are sorted according to a preset rule; The user is selected to receive a preset number of the first target advertisement information and / or the second target advertisement information, ranked at the top of the list.

13. A contextual online advertising delivery device, said device being used in a language learning application, characterized in that, include: A text acquisition unit is used to acquire text to be processed; wherein, the text to be processed is English text for language learning by the user, and the text to be processed includes multiple text units, wherein the text unit is a word; The target acquisition unit is used to acquire the target text unit and the target text in response to a user performing a context operation on the text to be processed. The user's context operation on the text to be processed includes a preceding operation and a following operation. The preceding operation expands the text content of the text to be processed by adding or replacing at least one text unit in the text to be processed. The following operation deletes the text content of the text to be processed by deleting at least one text unit in the text to be processed. A data acquisition unit is used to acquire a word-document correspondence table and a candidate sentence library; wherein, the word-document correspondence table includes multiple candidate words and advertising information associated with the candidate words, and the candidate sentence library includes multiple candidate sentences and advertising information associated with the candidate sentences; The first information acquisition unit is used to acquire first target advertising information based on the target text unit and the word document correspondence table; The second information acquisition unit is configured to acquire second target advertising information based on the target text and the candidate statement library; and An information providing unit is configured to provide the user with the target text and / or the target text unit, as well as the first target advertising information and / or the second target advertising information; The target acquisition unit includes: a syntax structure acquisition module, a weight level acquisition module, a first target text unit determination module, a target simplified text generation module, and a first data acquisition module. The grammar structure acquisition module is used to respond to the user's subsequent operation on the text to be processed, input the text to be processed into the grammar analysis model, and obtain the target grammar structure that matches the text to be processed; The weight level acquisition module is used to acquire the weight levels corresponding to multiple text units in the text to be processed according to the target syntax structure. The first target text unit determination module is used to determine a first target text unit among a plurality of text units when the weight level includes at least two levels; The target simplified text generation module is used to delete the first target text unit in the text to be processed to generate target simplified text; The first data acquisition module is used to take the first target text unit as the target text unit and the target simplified text as the target text; The information providing unit is further configured to: The target text is input into the syntax analysis model, and if a matching syntax structure exists, the target text is provided to the user. If no matching grammatical structure exists, a context operation error will be displayed and the program will exit.

14. A server, characterized in that, include: processor; Memory used to store the processor's executable instructions; The processor is configured to execute the instructions to implement the method as described in any one of claims 1 to 12.

15. A storage medium, characterized in that, When the instructions in the storage medium are executed by the processor of the electronic device, the electronic device is able to perform the method as described in any one of claims 1 to 12.

16. A computer program product comprising a computer program that, when executed by a processor, implements the method according to any one of claims 1 to 12.

Images