Methods, apparatus, electronic devices and storage media for determining object summaries

Abstract

This disclosure relates to a method, apparatus, electronic device, and storage medium for determining object summaries. The method includes: acquiring textual description information of an object; extracting key information and generating a summary from the textual description information according to a textual summarization model to obtain the object's summary information; the textual summarization model includes a key information extraction model and a summary generation model; wherein, the key information extraction model is trained based on the acquired textual description information of historical search objects and multiple search terms corresponding to the historical search objects; the summary generation model is trained based on the key textual information of historical search objects and filtered search terms; the key textual information is extracted from the textual description information using the key information extraction model; and the filtered search terms are obtained by filtering multiple search terms based on the key textual information. Thus, this disclosure can quickly and accurately generate summary information of an object.

Description

Technical Field

[0001] This disclosure relates to the field of Internet technology, and in particular to methods, apparatus, electronic devices and storage media for determining object summaries. Background Art

[0002] As a new information carrier, short videos contain not only image information but also textual descriptions, such as text information related to the content displayed in the short video edited by the user when uploading the video, as well as the short video title. Currently, extracting summary information from the textual descriptions of these short videos and applying the extracted summary information to search scenarios can help improve search accuracy.

[0003] However, current methods for extracting summaries from textual descriptions rely heavily on the quality of manually labeled datasets. Given the sheer volume of short videos, it is impossible to label enough data manually. Therefore, existing methods cannot generate ideal summaries. Summary of the Invention

[0004] This disclosure provides a method, apparatus, electronic device, and storage medium for determining object summaries. The technical solution of this disclosure is as follows:

[0005] According to a first aspect of the present disclosure, a method for determining an object digest is provided, comprising:

[0006] Get the text description information of the object;

[0007] Based on the text summarization model, key information is extracted and a summary is generated from the text description information to obtain the summary information of the object; the text summarization model includes a key information extraction model and a summary generation model.

[0008] The key information extraction model is trained based on the text description information of the historical search objects and multiple search terms corresponding to the historical search objects; the summary generation model is trained based on the key text information of the historical search objects and the filtered search terms; the key text information is extracted from the text description information by the key information extraction model; and the filtered search terms are obtained by filtering multiple search terms based on the key text information.

[0009] In some possible embodiments, the method for generating the key information extraction model includes:

[0010] Retrieve the text description information of historical search objects and multiple search terms corresponding to historical search objects;

[0011] Based on the text description information and multiple search terms, the first preset machine learning model is trained to extract key information, thus obtaining the key information extraction model.

[0012] In some possible embodiments, the method for generating the summary generation model includes:

[0013] Based on the key information extraction model, the text description information is extracted to obtain the key text information of the historical search objects;

[0014] Multiple search terms are filtered based on key text information to obtain the filtered search terms;

[0015] Based on key text information and filtered search terms, the second preset machine learning model is trained to generate a summary model.

[0016] In some possible embodiments, obtaining the text description information of the historical search object and multiple search terms corresponding to the historical search object includes:

[0017] Retrieve search logs; search logs include multiple historical search objects, text descriptions of each historical search object, and multiple search terms corresponding to each historical search object.

[0018] A preliminary screening of multiple search terms corresponding to each historical search object is performed to obtain the preliminary screened search terms.

[0019] In some possible embodiments, a preliminary screening is performed on multiple search terms corresponding to each historical search object to obtain preliminary screened search terms, including:

[0020] Each search term in the multiple search terms corresponding to each historical search object is segmented to obtain the segmentation results;

[0021] Determine the word frequency of each word in the word segmentation result;

[0022] Sort the words according to their frequency.

[0023] Based on the sorting results of each word, the target word is obtained;

[0024] Search terms containing the target term are selected from multiple search terms corresponding to each historical search object and used as the initial filtered search terms.

[0025] In some possible embodiments, a key information extraction model is trained on a first preset machine learning model based on text description information and multiple search terms to obtain a key information extraction model, including:

[0026] Using textual description information as input and each search term from multiple search terms as output labels, the first preset machine learning model is trained to extract key information, thus obtaining the key information extraction model.

[0027] In some possible embodiments, multiple search terms are filtered based on key text information to obtain filtered search terms, including:

[0028] Determine the matching value between each search term and key text information among multiple search terms;

[0029] Search terms with a match value greater than or equal to a preset match value will be used as the filtered search terms.

[0030] In some possible embodiments, a second preset machine learning model is trained for summarization generation based on key text information and filtered search terms to obtain a summarization generation model, including:

[0031] Using key text information as input and filtered search terms as output, the second preset machine learning model is trained to generate a summary model.

[0032] According to a second aspect of the present disclosure, an object summary determination apparatus is provided, comprising:

[0033] The acquisition module is configured to retrieve the text description information of the object.

[0034] The summary generation module is configured to extract key information from text description information and generate a summary of the object based on a text summarization model; the text summarization model includes a key information extraction model and a summary generation model.

[0035] The key information extraction model is trained based on the text description information of the historical search objects and multiple search terms corresponding to the historical search objects; the summary generation model is trained based on the key text information of the historical search objects and the filtered search terms; the key text information is extracted from the text description information by the key information extraction model; and the filtered search terms are obtained by filtering multiple search terms based on the key text information.

[0036] In some possible embodiments, it also includes:

[0037] The first model generation module is configured to obtain the text description information of historical search objects and multiple search terms corresponding to the historical search objects; and to perform key information extraction training on the first preset machine learning model based on the text description information and multiple search terms to obtain the key information extraction model.

[0038] In some possible embodiments, it also includes:

[0039] The second model generation module is configured to extract text description information based on the key information extraction model to obtain key text information of historical search objects; filter multiple search terms based on the key text information to obtain filtered search terms; and train the second preset machine learning model to generate a summary based on the key text information and the filtered search terms to obtain a summary generation model.

[0040] In some possible embodiments, the first model generation module is further configured to perform the following: obtaining search logs; the search logs include multiple historical search objects, textual description information of each historical search object, and multiple search terms corresponding to each historical search object; and performing preliminary filtering on the multiple search terms corresponding to each historical search object to obtain preliminary filtered search terms.

[0041] In some possible embodiments, the first model generation module is further configured to perform word segmentation processing on each of the multiple search terms corresponding to each historical search object to obtain word segmentation results; determine the word frequency of each word in the word segmentation results; sort each word according to the word frequency; obtain the target word according to the sorting results of each word; and take the search terms containing the target word from the multiple search terms corresponding to each historical search object as the search terms after preliminary screening.

[0042] In some possible embodiments, the first model generation module is further configured to perform key information extraction training on the first preset machine learning model by taking text description information as input and taking each of the multiple search terms as output labels in sequence, to obtain a key information extraction model.

[0043] In some possible embodiments, the second model generation module is further configured to determine the matching value between each search term and the key text information among a plurality of search terms; and to use search terms with matching values ​​greater than or equal to preset matching values ​​as filtered search terms.

[0044] In some possible embodiments, the second model generation module is further configured to perform summary generation training on a second preset machine learning model by taking key text information as input and the filtered search terms as output, to obtain a summary generation model.

[0045] According to a third aspect of the present disclosure, an electronic device is provided, comprising:

[0046] processor;

[0047] Memory used to store processor-executable instructions;

[0048] The processor is configured to execute instructions to implement the object digest determination method provided in the first aspect of the present disclosure.

[0049] According to a fourth aspect of the present disclosure, a computer-readable storage medium is provided, which, when executed by a processor of an electronic device, enables the electronic device to perform the object digest determination method provided in the first aspect of the present disclosure.

[0050] According to a fifth aspect of the present disclosure, a computer program product is provided, comprising a computer program stored in a readable storage medium, wherein at least one processor of a computer device reads from the readable storage medium and executes the computer program, causing the computer device to perform the object digest determination method provided in the first aspect of the present disclosure.

[0051] The technical solutions provided by the embodiments of this disclosure have at least the following beneficial effects:

[0052] By acquiring the textual description information of an object, and then extracting key information and generating a summary based on the textual description information using a text summarization model, the object's summary information is obtained. The text summarization model includes a key information extraction model and a summary generation model. The key information extraction model is trained based on the acquired textual description information of historical search objects and multiple search terms corresponding to those objects. The summary generation model is trained based on the key textual information of historical search objects and filtered search terms. The key textual information is extracted from the textual description information using the key information extraction model. The filtered search terms are obtained by filtering multiple search terms based on the key textual information. In this way, the object's summary information can be generated quickly and accurately.

[0053] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the disclosure. Attached Figure Description

[0054] 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.

[0055] Figure 1 This is a schematic diagram illustrating an application environment according to an exemplary embodiment;

[0056] Figure 2 This is a flowchart illustrating an object summary determination method according to an exemplary embodiment;

[0057] Figure 3 This is a flowchart illustrating a method for generating a key information extraction model according to an exemplary embodiment;

[0058] Figure 4This is a flowchart illustrating an exemplary embodiment of obtaining textual description information of a historical search object and multiple search terms corresponding to the historical search object;

[0059] Figure 5 This is a flowchart illustrating a preliminary screening of multiple search terms corresponding to each historical search object, according to an exemplary embodiment.

[0060] Figure 6 This is a flowchart illustrating a method for generating a summary generation model according to an exemplary embodiment;

[0061] Figure 7 This is a flowchart illustrating a method for filtering multiple search terms based on key text information, according to an exemplary embodiment.

[0062] Figure 8 This is a block diagram illustrating an object summary determination apparatus according to an exemplary embodiment;

[0063] Figure 9 This is a block diagram illustrating an electronic device for object summary determination according to an exemplary embodiment. Detailed Implementation

[0064] 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.

[0065] 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 first 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 this disclosure 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.

[0066] It should be noted that the user information involved in this disclosure (including but not limited to user device information, user personal information, etc.) is all information authorized by the user or fully authorized by all parties.

[0067] Please see Figure 1 , Figure 1 This is a schematic diagram illustrating an application environment according to an exemplary embodiment, such as... Figure 1 As shown, the application environment may include server 01 and client 02; server 01 and client 02 are connected wirelessly or via wired connection.

[0068] First, server 01 generates a text summary model. Then, based on the text summary model, it extracts key information and generates a summary of the obtained text description information of the object to obtain the object's summary information. Subsequently, server 01 can push the object's summary information to client 02 for display on a specified page of client 02.

[0069] The server 01 can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN (Content Delivery Network), and big data and artificial intelligence platforms. The operating system running on the server can include, but is not limited to, Android, iOS, Linux, Windows, Unix, etc.

[0070] The client 02 can be, but is not limited to, smartphones, desktop computers, tablets, laptops, smart speakers, digital assistants, augmented reality (AR) / virtual reality (VR) devices, smart wearable devices, etc. It can also be software running on the aforementioned clients, such as applications or mini-programs. Optionally, the operating system running on the client can be, but is not limited to, Android, iOS, Linux, Windows, Unix, etc.

[0071] In addition, it should be noted that, Figure 1 The example shown is merely one application environment of the object summary determination method provided in this disclosure. In practical applications, other application environments may also be included.

[0072] The following describes an embodiment of the object summary determination method provided in this disclosure. Figure 2 This is a flowchart illustrating an object summary determination method according to an exemplary embodiment, such as... Figure 2 As shown, the object digest determination method can be applied to the server side and includes the following steps:

[0073] In step S201, the text description information of the object is obtained.

[0074] The object is content that matches the actual application scenario; in some possible application scenarios, the object may include music, short videos, products, news information, advertisements, etc.

[0075] In step S203, key information is extracted and a summary is generated from the text description information according to the text summarization model to obtain the summary information of the object; the text summarization model includes a key information extraction model and a summary generation model.

[0076] In practical applications, the text description information of an object is input into the text summarization model. The text summarization model first extracts key information from the text description information through the key information extraction model. The key information extraction model outputs the key text information of the object and passes it to the input of the summary generation model. Then, the summary generation model generates a summary from the key text information. Finally, the output of the text summarization model is the summary information output by the summary generation model.

[0077] In this embodiment of the disclosure, the key information extraction model is trained based on the obtained text description information of historical search objects and multiple search terms corresponding to the historical search objects; the summary generation model is trained based on the key text information of historical search objects and the filtered search terms; wherein, the key text information is extracted from the text description information by the key information extraction model; and the filtered search terms are obtained by filtering multiple search terms based on the key text information.

[0078] The generation methods for the key information extraction model and the summary generation model are described below. The key information extraction model and the summary generation model are trained separately. In practical applications, the information of the object to be summarized is first input into the key information extraction model, and the output is then input into the summary generation model to output the corresponding summary information.

[0079] Figure 3 This is a flowchart illustrating a method for generating a key information extraction model according to an exemplary embodiment, such as... Figure 3 As shown, the following steps are included:

[0080] In step S301, the text description information of the historical search object and multiple search terms corresponding to the historical search object are obtained.

[0081] In this embodiment of the disclosure, a key information extraction model is trained through steps S301 to S303. During the training of the key information extraction model, a training sample dataset is first constructed.

[0082] In related fields, if we want to train a model to determine the corresponding text summary for a specific object, we must have an labeled dataset. However, there is still no large-scale Chinese dataset that is both general-purpose and accurately describes the relevant domain. For example, in the field of short videos, generating corresponding text summaries for short video text requires not only a large amount of short video text, but also the corresponding summary of the short video as an annotation. Currently, such datasets are still constructed manually, but the scale of such datasets is large, and constructing them manually is very labor-intensive and time-consuming.

[0083] Based on this, in this embodiment of the disclosure, the server can automatically construct the required training sample dataset by obtaining the text description information of historical search objects and multiple search terms corresponding to the historical search objects. The aforementioned historical search objects represent objects that the user has previously searched for; these objects are content that matches the actual application scenario. In some possible application scenarios, objects may include music, short videos, products, news information, advertisements, etc. The aforementioned text description information includes Chinese descriptive text about the content displayed by the object. The aforementioned multiple search terms corresponding to the historical search objects are the search terms used by the user during past searches; generally, different users or the same user can use multiple different search terms to search for the same object, so one object can correspond to multiple different search terms.

[0084] In some possible embodiments, the above-described acquisition of the text description information of the historical search object and the multiple search terms corresponding to the historical search object may include, for example: Figure 4 The following steps are shown:

[0085] In step S401, the search log is obtained; the search log includes multiple historical search objects, text description information of each historical search object, and multiple search terms corresponding to each historical search object.

[0086] In this step, the server stores search logs for objects, which include log data for multiple historical search objects. By retrieving the search logs, textual descriptions of each historical search object and multiple search terms corresponding to each historical search object can be obtained from the search logs.

[0087] In step S403, multiple search terms corresponding to each historical search object are initially filtered to obtain the preliminarily filtered search terms.

[0088] Since search terms need to be used as annotations during training, and the number of search terms for each historical search object obtained directly from the search log is large, some of which may be search terms that do not match the actual content of the historical search object, meaning that the search term cannot accurately describe the historical search object; if the training sample dataset is built directly based on the data in the original search log, the dataset will contain a lot of noise, which will ultimately affect the model's generation performance.

[0089] Therefore, in the above embodiments, by performing preliminary screening on multiple search terms corresponding to each historical search object and constructing a training sample dataset based on the preliminary screening search terms, the accuracy of the dataset can be improved, thereby reducing the noise impact on the model during subsequent training and ultimately improving the accuracy of the model output results.

[0090] In some possible embodiments, the preliminary filtering of multiple search terms corresponding to each historical search object to obtain the preliminary filtered search terms may include, for example: Figure 5 The following steps are shown:

[0091] In step S501, each search term in the multiple search terms corresponding to each historical search object is segmented to obtain the segmentation result.

[0092] In this step, the server can use a word segmentation algorithm to segment each search term among multiple search terms corresponding to each historical search object, and obtain the word segmentation result corresponding to each search term; wherein, the word segmentation algorithm can include a word segmentation algorithm based on string matching or a word segmentation algorithm based on statistics and machine learning.

[0093] In step S503, the word frequency of each word in the word segmentation result is determined.

[0094] In step S505, the words are sorted according to their frequency.

[0095] In step S507, the target word is obtained based on the sorting results of each word.

[0096] In steps S503 to S507 above, the word segmentation result corresponding to each search term includes at least one word; different search terms may contain the same word after word segmentation. All words obtained after word segmentation of all search terms are counted, the frequency of each word in all words is determined, and the word frequency of each word is obtained; then, according to the word frequency of each word, each word is sorted, and according to the sorting result, the top N words can be used as target words.

[0097] In step S509, the search terms containing the target term in the multiple search terms corresponding to each historical search object are used as the search terms after preliminary screening.

[0098] In this step, multiple search terms corresponding to the historical search object are initially filtered based on the target term, that is, search terms that do not contain the target term are deleted, and search terms that do contain the target term are retained.

[0099] The above embodiments are illustrated below with a specific example. Suppose that in a short video application scenario, a certain historical search object is short video A, and the user's historical search terms include six search terms: abc, ad, cd, de, a, and b. First, these six search terms are segmented into words, resulting in the following: abc: a, b, c; ad: a, d; cd: c, d; de: d, e; a: a; and b: b. Then, the word frequencies of each word are determined as: a (3 times), b (2 times), c (2 times), d (3 times), and e (1 time). The corresponding ranking is: a, d, b, c, e. Words with the same frequency are randomly ordered. The top two ranked words are selected as target words, so a and d are the target words. After initial screening, search term b is deleted, leaving the remaining search terms: abc, ad, cd, de, and a.

[0100] In the above embodiments, by performing word segmentation on each search term for each historical search object, the most frequently occurring word is counted based on the word segmentation results. This most frequently occurring word is the word that users use more often, and it is more likely to accurately describe the object. Thus, based on the most frequently occurring word, multiple search terms are initially filtered, inaccurate search terms can be deleted, and search terms that match the actual description of the object can be obtained. This is beneficial to improving the quality of the training sample dataset.

[0101] In step S303, the first preset machine learning model is trained to extract key information based on the text description information and multiple search terms, thereby obtaining the key information extraction model.

[0102] In this embodiment of the disclosure, during the training of the key information extraction model, it is also necessary to build a first preset machine learning model, and use the training sample dataset constructed in the above steps to train the first preset machine learning model. When the preset training termination condition is met, the trained key information extraction model is obtained.

[0103] In some possible embodiments, the above-described process of training a first preset machine learning model to extract key information based on text description information and multiple search terms, to obtain a key information extraction model, may include:

[0104] Using textual description information as input and each search term from multiple search terms as output labels, the first preset machine learning model is trained to extract key information, thus obtaining the key information extraction model.

[0105] The constructed training sample dataset includes multiple sample pairs. Each sample pair includes textual descriptions of historical search objects and any one of multiple search terms. Typically, one object corresponds to multiple search terms, and each search term can form a sample pair with the textual descriptions. Therefore, during training, the textual descriptions of historical search objects are used as input, and the corresponding search terms are used as labels to train the first preset machine learning model for key information extraction. Specifically, for the search terms in the current sample pair, they are matched sequentially with words in the textual descriptions. If a match is successful, the corresponding textual information is extracted. This corresponding textual information includes words with similar or identical meanings to the search terms. The corresponding textual information extracted based on the search terms, i.e., the key textual information in the textual descriptions, can also be referred to as a refined version of the textual descriptions. When the preset training termination conditions are met, the trained key information extraction model is obtained.

[0106] The preset training termination conditions may include ending training when the number of iterations reaches a preset number; or ending training when the model parameters reach the optimization target.

[0107] In the above embodiments, a key information extraction model is obtained by training the constructed training sample dataset. The key information extraction model is used to extract key information from the original text description information of the object and remove irrelevant information to obtain key text information that is strongly related to the content displayed by the object, that is, a more concise description text. In this way, in the subsequent process of generating object summaries, more accurate object summary information can be obtained based on the more concise description text.

[0108] Figure 6 This is a flowchart illustrating a method for generating a summary generation model according to an exemplary embodiment, such as... Figure 6 As shown, the following steps are included:

[0109] In step S601, the text description information is extracted according to the key information extraction model to obtain the key text information of the historical search object.

[0110] In step S603, multiple search terms are filtered based on key text information to obtain the filtered search terms.

[0111] In this embodiment of the disclosure, after the server completes the training of the key information extraction model, it uses the trained key information extraction model to extract text description information to obtain key text information of historical search objects; then, based on the key text information, it filters multiple search terms to obtain the filtered search terms.

[0112] In the above embodiments, multiple search terms can be multiple search terms that have already undergone preliminary screening; in the above steps S601 to S603, the key text information of historical search objects is used to further screen the multiple search terms that have undergone preliminary screening, so as to further improve the quality of the dataset.

[0113] In some possible embodiments, the above-described filtering of multiple search terms based on key text information to obtain filtered search terms may include, for example: Figure 7 The following steps are shown:

[0114] In step S701, the matching value between each search term and the key text information is determined among multiple search terms.

[0115] In step S703, search terms with matching values ​​greater than or equal to preset matching values ​​are selected as filtered search terms.

[0116] Specifically, the server can calculate the similarity value between each search term and the key text information, and use this similarity value as the matching value. In short video applications, the extracted key text information is usually in phrase form, and the search terms are also in phrase form. Therefore, the similarity value can be determined by calculating the number of identical characters between each search term and the key text information; the more identical characters, the higher the similarity value and the higher the matching value. The preset matching value can be determined according to actual needs.

[0117] In the above embodiments, by determining the matching value between each search term and the key text information among multiple search terms, a secondary filtering of search terms can be achieved. The search terms obtained by the final filtering can accurately describe the actual content displayed by the object, which can further improve the quality of the dataset and further reduce the impact of noise.

[0118] In addition, to ensure the accuracy of the program's filtering results, the filtering results can be sampled and checked manually to determine whether the search terms of each historical search object are reliable. If they are not reliable, repeat the above steps S303, S601 to S603, and execute the subsequent steps after determining that they are reliable.

[0119] In step S605, the second preset machine learning model is trained to generate a summary based on the key text information and the filtered search terms, thus obtaining the summary generation model.

[0120] In this embodiment of the disclosure, a summary generation model is trained through step S605. During the training of the summary generation model, a second preset machine learning model is built. Based on the selected search terms and key text information, the second preset machine learning model is trained to generate summaries. When the preset training termination condition is met, the trained summary generation model is obtained.

[0121] In some possible embodiments, the above-described process of training a second preset machine learning model to generate a summary based on key text information and filtered search terms may include:

[0122] Using key text information as input and filtered search terms as output, the second preset machine learning model is trained to generate a summary model.

[0123] During the training process, the key text information of the historical search objects extracted by the key information extraction model is used as the input of the second preset machine learning model. The second preset machine learning model is trained with any search term after output filtering as the target. When the preset training termination condition is met, the trained summary generation model is obtained.

[0124] The preset training termination conditions may include ending training when the number of iterations reaches a preset number; or ending training when the model parameters reach the optimization target.

[0125] In the above embodiments, using key text information of historical search objects as input, compared to directly using raw text description information as input, allows the model to be less affected by noise during training, thus generating more accurate summaries at this stage. Furthermore, the final output of the summary generation model is a phrase similar to the search terms, which can be more widely used in object search scenarios.

[0126] In this embodiment of the disclosure, a key information extraction model is obtained by training through the above steps S301 to S303, and a summary generation model is obtained by training through steps S601 to S605. The key information extraction model and the summary generation model constitute the text summarization model described in this disclosure.

[0127] This disclosure provides a method for generating a text summarization model. Specifically, it obtains a key information extraction model based on steps S301-S303, a summary generation model based on steps S601-S605, and finally obtains a text summarization model based on the key information extraction model and the summary generation model. This method solves the problem of not being able to quickly and effectively generate accurate text summarization models due to the lack of large-scale Chinese datasets with general domains or precise descriptions. Furthermore, this disclosure not only allows for the rapid and effective generation of accurate text summarization models, but the generated text summarization models can also be used to determine object summaries.

[0128] In this embodiment of the disclosure, the final summary information can very accurately describe the actual content displayed by the object. As mentioned above, this summary information can be more widely used in object search scenarios; for example, the summary information can be used on a specified page on the client to intelligently complete the content entered by the user in the search box, so that the user can directly select the intelligently completed terms. Since the intelligently completed terms, i.e., the summary information, are generated based on existing objects, the user can definitely obtain objects strongly related to the selected terms, thus improving the search success rate and enhancing the user experience; another example is that the summary information can be displayed on the display interface of the corresponding object to help users quickly understand the content displayed by the object.

[0129] Figure 8 This is a block diagram illustrating an object summary determination apparatus according to an exemplary embodiment. (Refer to...) Figure 8 The device includes an acquisition module 801 and a summary generation module 802;

[0130] Module 801 is configured to retrieve the text description information of the object.

[0131] The summary generation module 802 is configured to perform key information extraction and summary generation on the text description information according to the text summary model to obtain the summary information of the object; the text summary model includes a key information extraction model and a summary generation model.

[0132] The key information extraction model is trained based on the text description information of the historical search objects and multiple search terms corresponding to the historical search objects; the summary generation model is trained based on the key text information of the historical search objects and the filtered search terms; the key text information is extracted from the text description information by the key information extraction model; and the filtered search terms are obtained by filtering multiple search terms based on the key text information.

[0133] In some possible embodiments, it also includes:

[0134] The first model generation module is configured to obtain the text description information of historical search objects and multiple search terms corresponding to the historical search objects; and to perform key information extraction training on the first preset machine learning model based on the text description information and multiple search terms to obtain the key information extraction model.

[0135] In some possible embodiments, it also includes:

[0136] The second model generation module is configured to extract text description information based on the key information extraction model to obtain key text information of historical search objects; filter multiple search terms based on the key text information to obtain filtered search terms; and train the second preset machine learning model to generate a summary based on the key text information and the filtered search terms to obtain a summary generation model.

[0137] In some possible embodiments, the first model generation module is further configured to perform the following: obtaining search logs; the search logs include multiple historical search objects, textual description information of each historical search object, and multiple search terms corresponding to each historical search object; and performing preliminary filtering on the multiple search terms corresponding to each historical search object to obtain preliminary filtered search terms.

[0138] In some possible embodiments, the first model generation module is further configured to perform word segmentation processing on each of the multiple search terms corresponding to each historical search object to obtain word segmentation results; determine the word frequency of each word in the word segmentation results; sort each word according to the word frequency; obtain the target word according to the sorting results of each word; and take the search terms containing the target word from the multiple search terms corresponding to each historical search object as the search terms after preliminary screening.

[0139] In some possible embodiments, the first model generation module is further configured to perform key information extraction training on the first preset machine learning model by taking text description information as input and taking each of the multiple search terms as output labels in sequence, to obtain a key information extraction model.

[0140] In some possible embodiments, the second model generation module is further configured to determine the matching value between each search term and the key text information among a plurality of search terms; and to use search terms with matching values ​​greater than or equal to preset matching values ​​as filtered search terms.

[0141] In some possible embodiments, the second model generation module is further configured to perform summary generation training on a second preset machine learning model by taking key text information as input and the filtered search terms as output, to obtain a summary generation model.

[0142] 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.

[0143] Figure 9 This is a block diagram illustrating an electronic device for object summary determination according to an exemplary embodiment.

[0144] This electronic device can be a server or a terminal device, and its internal structure diagram can be as follows: Figure 9 As shown, the electronic device includes a processor, memory, and a network interface connected via a system bus. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The network interface is used to communicate with external terminals via a network connection. When executed by the processor, the computer program implements an object summary determination method.

[0145] Those skilled in the art will understand that Figure 9 The structure shown is merely a block diagram of a portion of the structure related to the present disclosure and does not constitute a limitation on the electronic device to which the present disclosure is applied. A specific electronic device may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0146] In an exemplary embodiment, an electronic device is also provided, including: a processor; and a memory for storing processor-executable instructions; wherein the processor is configured to execute the instructions to implement the object digest determination method as described in the embodiments of this disclosure.

[0147] In an exemplary embodiment, a computer-readable storage medium is also provided, which, when the instructions in the computer-readable storage medium are executed by a processor of an electronic device, enables the electronic device to perform the object digest determination method of the embodiments of this disclosure.

[0148] In an exemplary embodiment, a computer program product is also provided, comprising a computer program stored in a readable storage medium, wherein at least one processor of a computer device reads from the readable storage medium and executes the computer program, causing the computer device to perform the object digest determination method of the present disclosure embodiments.

[0149] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. This computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and RAMbus dynamic RAM (RDRAM), etc.

[0150] 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 application 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.

[0151] 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 method for determining object summaries, characterized in that, include: Get the text description information of the object; Based on a text summarization model, key information is extracted and a summary is generated from the text description information to obtain the summary information of the object; the text summarization model includes a key information extraction model and a summary generation model. The key information extraction model is trained by taking the textual description information of the historical search objects as input and multiple search terms corresponding to the historical search objects as output labels; the summary generation model is trained based on the key textual information of the historical search objects and the filtered search terms. The key information extraction model is used to extract key text information from the text description information based on the multiple search terms; the filtered search terms are obtained by filtering the multiple search terms based on the key text information.

2. The object digest determination method according to claim 1, characterized in that, The method for generating the key information extraction model includes: Obtain the text description information of the historical search object and multiple search terms corresponding to the historical search object; Based on the text description information and the multiple search terms, the first preset machine learning model is trained to extract key information, thereby obtaining the key information extraction model.

3. The object digest determination method according to claim 2, characterized in that, The method for generating the summary generation model includes: Based on the key information extraction model, the text description information is extracted to obtain the key text information of the historical search object; Based on the key text information, the multiple search terms are filtered to obtain the filtered search terms; Based on the key text information and the filtered search terms, the second preset machine learning model is trained to generate a summary, thereby obtaining the summary generation model.

4. The object digest determination method according to claim 2, characterized in that, The step of obtaining the text description information of the historical search object and the multiple search terms corresponding to the historical search object includes: Obtain the search log; the search log includes multiple historical search objects, text description information of each historical search object, and multiple search terms corresponding to each historical search object. A preliminary screening of multiple search terms corresponding to each historical search object is performed to obtain the preliminary screened search terms.

5. The object digest determination method according to claim 4, characterized in that, The preliminary filtering of multiple search terms corresponding to each historical search object to obtain the preliminary filtered search terms includes: Each search term among the multiple search terms corresponding to each historical search object is segmented to obtain the segmentation result; Determine the word frequency of each word in the word segmentation result; The words are sorted according to their frequency. Based on the sorting results of the various words, the target word is obtained; The search terms containing the target term are selected from the search terms corresponding to each historical search object and used as the search terms after preliminary filtering.

6. The object digest determination method according to claim 2, characterized in that, The step of training a first preset machine learning model to extract key information based on the text description information and the multiple search terms, to obtain the key information extraction model, includes: Using the text description information as input, and each of the multiple search terms as output labels in sequence, the first preset machine learning model is trained to extract key information, thereby obtaining the key information extraction model.

7. The object digest determination method according to claim 3, characterized in that, The process of filtering the multiple search terms based on the key text information to obtain the filtered search terms includes: Determine the matching value between each of the multiple search terms and the key text information; Search terms whose matching value is greater than or equal to a preset matching value are used as the filtered search terms.

8. The object digest determination method according to claim 3, characterized in that, The step of training a second preset machine learning model to generate a summary based on the key text information and the filtered search terms, to obtain the summary generation model, includes: Using the key text information as input and the filtered search terms as output, the second preset machine learning model is trained to generate a summary model.

9. An object summary determination device, characterized in that, include: The acquisition module is configured to retrieve the text description information of the object. The summary generation module is configured to perform key information extraction and summary generation on the text description information according to the text summarization model to obtain the summary information of the object; the text summarization model includes a key information extraction model and a summary generation model. The key information extraction model is trained by taking the textual description information of the historical search objects as input and multiple search terms corresponding to the historical search objects as output labels; the summary generation model is trained based on the key textual information of the historical search objects and the filtered search terms. The key information extraction model is used to extract key text information from the text description information based on the multiple search terms; the filtered search terms are obtained by filtering the multiple search terms based on the key text information.

10. The object summary determination apparatus according to claim 9, characterized in that, Also includes: The first model generation module is configured to obtain the text description information of the historical search object and multiple search terms corresponding to the historical search object; Based on the text description information and the multiple search terms, the first preset machine learning model is trained to extract key information, thereby obtaining the key information extraction model.

11. The object summary determination apparatus according to claim 10, characterized in that, Also includes: The second model generation module is configured to extract the text description information based on the key information extraction model to obtain the key text information of the historical search object. Based on the key text information, the multiple search terms are filtered to obtain the filtered search terms; Based on the key text information and the filtered search terms, the second preset machine learning model is trained to generate a summary, thereby obtaining the summary generation model.

12. The object summary determination apparatus according to claim 10, characterized in that, The first model generation module is further configured to perform the acquisition of search logs; the search logs include multiple historical search objects, text description information of each historical search object, and multiple search terms corresponding to each historical search object; and perform preliminary screening on the multiple search terms corresponding to each historical search object to obtain preliminary screened search terms.

13. The object summary determination apparatus according to claim 12, characterized in that, The first model generation module is further configured to perform word segmentation processing on each of the multiple search terms corresponding to each historical search object to obtain word segmentation results; and determine the word frequency of each word in the word segmentation results. Based on the word frequency of each word, the words are sorted; based on the sorting results of each word, the target word is obtained; the search terms containing the target word in the multiple search terms corresponding to each historical search object are used as the search terms after preliminary screening.

14. The object summary determination apparatus according to claim 10, characterized in that, The first model generation module is further configured to take the text description information as input, take each of the multiple search terms as output labels in sequence, and train the first preset machine learning model to extract key information, thereby obtaining the key information extraction model.

15. The object summary determination apparatus according to claim 11, characterized in that, The second model generation module is further configured to determine the matching value between each search term among the plurality of search terms and the key text information; and to use the search terms whose matching value is greater than or equal to a preset matching value as the filtered search terms.

16. The object summary determination apparatus according to claim 11, characterized in that, The second model generation module is further configured to take the key text information as input and the filtered search terms as output to train the second preset machine learning model to generate a summary model.

17. An electronic device, 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 object digest determination method as described in any one of claims 1 to 8.

18. A computer-readable storage medium, characterized in that, When the instructions in the computer-readable storage medium are executed by the processor of the electronic device, the electronic device is able to perform the object digest determination method as described in any one of claims 1 to 8.

19. A computer program product, characterized in that, The computer program product includes a computer program stored in a readable storage medium, wherein at least one processor of a computer device reads from the readable storage medium and executes the computer program, causing the computer device to perform the object digest determination method as described in any one of claims 1 to 8.

Images