Content delivery method, related devices and media

Abstract

The present disclosure provides a content delivery method, related device and medium, which belongs to the technical field of big data. The content delivery method comprises: determining that a first condition is established based on the number of published contents of the same type as a first content to be delivered and the number of published contents of a first object for delivering the first content; in response to the first condition being established, determining a first keyword based on the first content, expanding a second keyword based on the first keyword, and integrating the first keyword and the second keyword into a target keyword; obtaining object labels of each object in an object set on a content delivery platform; selecting a target object from the object set based on the matching degree between the target keyword and the object label; and delivering the first content based on the selected target object. The present disclosure improves the object reach efficiency of unpopular content delivery and improves the network resource utilization rate. The present disclosure can be applied to various scenes such as information delivery, big data and artificial intelligence.

Description

Technical Field

[0001] This disclosure relates to the field of big data, and in particular to a content delivery method, related apparatus and medium. Background Technology

[0002] The internet involves a large amount of content push and delivery. Currently, content push and delivery generally adopts a method of finding popular content that matches the browsing history or interests of the target audience, i.e., "people finding things." This is extremely unfriendly to some new content creators or new types of content from established authors. Because they lack the accumulation of historical delivery data, the traditional "people finding things" method will allocate very little exposure to the content, which is detrimental to the efficiency of reaching the target audience for specific content. This forces some content to reach the target audience through additional channels, wasting network resources and reducing the utilization rate of network resources. Summary of the Invention

[0003] This disclosure provides a content delivery method, related apparatus, and medium that can improve the reach efficiency of delivering niche content and increase the utilization rate of network resources.

[0004] According to one aspect of this disclosure, a content delivery method is provided, comprising:

[0005] Based on the number of published content of the same type as the first content to be published and the number of published content of the first object to publish the first content, the first condition is determined to be met;

[0006] In response to the first condition being met, a first keyword is determined based on the first content, a second keyword is expanded based on the first keyword, and the first keyword and the second keyword are integrated as the target keyword;

[0007] Obtain the object tags of each object in the object set on the content delivery platform;

[0008] Based on the matching degree between the target keywords and the object tags, target objects are selected from the object set;

[0009] The first content is delivered based on the selected target object.

[0010] According to one aspect of this disclosure, a content delivery device is provided, comprising:

[0011] The condition determination unit is used to determine that the first condition is met based on the number of published content of the same type as the first content to be delivered and the number of published content of the first object delivering the first content.

[0012] A keyword determination unit is used to, in response to the first condition being met, determine a first keyword based on the first content, expand a second keyword based on the first keyword, and integrate the first keyword and the second keyword into a target keyword.

[0013] The tag acquisition unit is used to acquire the object tags of each object in the object set on the content delivery platform.

[0014] An object selection unit is used to select a target object from the object set based on the matching degree between the target keyword and the object tag;

[0015] The content delivery unit is used to deliver the first content based on the selected target object.

[0016] Optionally, the condition determination unit is specifically used for:

[0017] If the number of published contents of the first object that delivers the first content is less than the first threshold, then the first condition is determined to be met.

[0018] Determine the first type of the first content;

[0019] Determine the number of published content of the first type on the content delivery platform;

[0020] If the number of published content of the first type is less than the second threshold, then the first condition is determined to be met.

[0021] Optionally, the keyword determination unit is specifically used for:

[0022] Extract the first set of keywords from the text description of the first content;

[0023] Extract the second sub-keyword from the text description of the second content placed before the first content by the first object;

[0024] Based on the multimodal analysis of the first content, the third sub-keyword is extracted;

[0025] The first sub-keyword, the second sub-keyword, and the third sub-keyword are integrated into the first keyword.

[0026] Optionally, the keyword determination unit is specifically used for:

[0027] Input the first keyword into the related word prediction model to obtain the fourth sub-keyword;

[0028] The first keyword is expanded using a knowledge graph to obtain the fifth sub-keyword;

[0029] The fourth and fifth sub-keywords are combined into the second keyword.

[0030] Optionally, the target keyword has a first weight, the object tag has a second weight, and the object selection unit includes:

[0031] A matching degree determination unit is used to determine the matching degree between the target keyword and the object tag based on the target keyword, the first weight, the object tag, and the second weight;

[0032] The target object selection unit is used to select a target object from the object set based on the matching degree of each of the objects in the object set.

[0033] Optionally, the second weight includes a first sub-weight from the first source and a second sub-weight from the second source, and the matching degree determination unit is specifically used for:

[0034] Based on the target keyword, the first weight, the object tag, and the first sub-weight, a first matching degree corresponding to the first source is determined;

[0035] Based on the target keyword, the first weight, the object tag, and the second sub-weight, a second matching degree corresponding to the second source is determined;

[0036] The matching degree is determined based on the first matching degree and the second matching degree.

[0037] Optionally, the first weight of the first keyword is a fixed weight, and the content delivery device further includes a weight determination unit, specifically used for:

[0038] The first weight of the second keyword is determined, wherein the first weight of the fourth sub-keyword is the confidence probability of the fourth sub-keyword output by the related word prediction model; the first weight of the fifth sub-keyword is determined based on the distance between the fifth sub-keyword and the first keyword in the knowledge graph.

[0039] Optionally, the matching degree determination unit is specifically used for:

[0040] For each object in the object set, determine a target object tag that matches the target keyword from the object tags of that object;

[0041] For each target object tag, determine the product of the second weight of the target object tag and the first weight of the matching target keyword;

[0042] The matching degree is obtained by summing the products of each target object label.

[0043] Optionally, the target object selection unit is specifically used for:

[0044] In the object set, candidate objects with a matching degree greater than a third threshold are identified;

[0045] The candidate objects are sorted in descending order of matching degree;

[0046] The candidate objects ranked first in the sort are taken as the target objects.

[0047] Optionally, the tag acquisition unit is specifically used for:

[0048] For each object in the object set, obtain the set of content of interest represented by the object;

[0049] Obtain the third keyword of the content in the aforementioned content set;

[0050] For each of the third keywords, determine the number of contents in the content set that contain the third keyword;

[0051] If the number of contents is greater than the fourth threshold, the third keyword will be used as the object tag of the object.

[0052] Optionally, the object selection unit includes:

[0053] The first filtering unit is used to perform a first filtering on the object set based on the matching degree between the target keyword and the object tag, so as to obtain the first filtered objects;

[0054] A filtering condition acquisition unit is used to acquire the first filtering condition of the first object;

[0055] The second filtering unit is used to perform a second filtering on the first filtered objects based on the matching of the first filtering conditions and the attribute information of the first filtered objects, so as to obtain the target object.

[0056] Optionally, the second filtering unit is specifically used for:

[0057] Based on the matching of the first filtering condition with the attribute information of the first filtered object, a second filtering is performed on the first filtered object to obtain a second filtered object.

[0058] Obtain the attribute information of the second filtered object;

[0059] Display a distribution map of the objects after the second filtering based on the attribute information;

[0060] Display an attribute information checkbox to receive a second filtering condition input by the first object based on the distribution map;

[0061] Based on the matching of the second filtering conditions with the attribute information of the second filtered objects, a third filtering is performed on the second filtered objects to obtain the target object.

[0062] Optionally, the second filtering unit is specifically used for:

[0063] Based on the matching of the filtering conditions with the attribute information of the first filtered object, a second filtering is performed on the first filtered object to obtain a second filtered object.

[0064] The objects after the second filtering are sorted from high to low according to the matching degree;

[0065] The second filtered object in the top second position of the ranking is used as the first display object, and the first display object and the object tag of the first display object that matches the target keyword are displayed.

[0066] Select a portion of the second filtered objects from those that have never ranked in the first second place in the sorting, and display them as the second display objects, along with the object tags of the second display objects that match the target keyword.

[0067] The first object receives the number of objects to be retained, wherein the number of objects to be retained is selected by the first object based on the first display object, the object tags of the first display object that match the target keyword, the second display object, and the object tags of the second display object that match the target keyword.

[0068] The target object is obtained by retaining the second filtered object according to the stated retention number.

[0069] Optionally, the content delivery device further includes a first update unit, the first update unit being used for:

[0070] Obtain the interest representation of the target audience after the first content is delivered;

[0071] Based on the target object's interest representation, update the first weight of the target keyword in the first content, wherein the first weight is the ratio of a first number to a second number, the first number is the number of times the target object expressed interest after the first content containing the target keyword was delivered to the target object for the target keyword, and the second number is the number of times the first content containing the target keyword was delivered to the target object for the target keyword.

[0072] Optionally, the content delivery device further includes a second update unit, the second update unit being used for:

[0073] Obtain the interest representation of the target audience after the first content is delivered;

[0074] Based on the interest representation of the target object, update the second weight of the object tag of the target object, wherein the second weight is the ratio of the third number to the fourth number, the third number is the number of times the target object expressed interest after the first content was delivered to the target object with the object tag, and the fourth number is the number of times the first content was delivered to the target object with the object tag.

[0075] Optionally, the content delivery device further includes a display unit, which is used for:

[0076] Obtain the interest representation of the target audience after the first content is delivered;

[0077] Obtain the attribute information of the target object;

[0078] For each of the attribute information, determine the ratio of the target object corresponding to the attribute information that represents the interest;

[0079] The ratios corresponding to each of the aforementioned attribute information are displayed for filtering the delivery of the third content of the first object after the first content is delivered.

[0080] According to one aspect of this disclosure, an electronic device is provided, including a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the content delivery method described above.

[0081] According to one aspect of this disclosure, a computer-readable storage medium is provided, the storage medium storing a computer program that, when executed by a processor, implements the content delivery method described above.

[0082] According to one aspect of this disclosure, a computer program product is provided, the computer program product including a computer program that is read and executed by a processor of a computer device, causing the computer device to perform the content delivery method as described above.

[0083] In this embodiment, when the first condition is met based on the number of published content of the same type as the first content to be delivered and the number of published content of the first object to which the first content is delivered, it indicates that the author of the first content is relatively new or the type of the first content is relatively new. In this case, due to the lack of historical data accumulation, the first content will be allocated less exposure using the traditional model, reducing content reach efficiency. This embodiment determines a first keyword based on the first content and expands it to a second keyword. These are then integrated into a target keyword. Based on the matching degree between the target keyword and the object tags of various objects in the platform object set, a matching target object is selected for delivery to the first content. This delivery method does not rely on the accumulation of historical data for the first object, enabling rapid delivery to potentially interested target objects, improving the object reach efficiency of specific content, eliminating the need for additional channels, and improving network resource utilization. The automatic keyword generation method improves the automation level of content delivery, and the joint determination of the target object based on the basic first keyword and the expanded second keyword improves the accuracy of content delivery.

[0084] Other features and advantages of this disclosure will be set forth in the following description and will be apparent in part from the description or may be learned by practicing the disclosure. The objectives and other advantages of this disclosure may be realized and obtained by means of the structures particularly pointed out in the description, claims and drawings. Attached Figure Description

[0085] The accompanying drawings are provided to further understand the technical solutions of this disclosure and constitute a part of the specification. They are used together with the embodiments of this disclosure to explain the technical solutions of this disclosure and do not constitute a limitation on the technical solutions of this disclosure.

[0086] Figure 1 This is a schematic diagram of the system architecture for the content delivery method of this disclosure embodiment.

[0087] Figure 2A -K is a schematic diagram illustrating the application of the content delivery method according to the embodiments of this disclosure in a video delivery scenario;

[0088] Figure 3 This is a flowchart of a content delivery method according to an embodiment of the present disclosure;

[0089] Figure 4 It shows Figure 3 The flowchart for step 310, determining that the first condition is met;

[0090] Figure 5 It shows Figure 3 The flowchart for step 320, determining the first keyword based on the first content;

[0091] Figure 6 It shows Figure 5 A schematic diagram illustrating the process of extracting the third sub-keyword based on multimodal parsing of the first content;

[0092] Figure 7 It shows Figure 3 Step 320 is a flowchart of expanding the second keyword based on the first keyword;

[0093] Figure 8A -B shows Figure 7 A schematic diagram illustrating the process of expanding a second keyword based on a first keyword;

[0094] Figure 9 It shows Figure 3 A diagram illustrating the keyword list formed by integrating target keywords in step 320;

[0095] Figure 10 It shows in Figure 3 A flowchart for adjusting target keywords after step 320;

[0096] Figure 11 It shows Figure 3 The flowchart for step 330, obtaining the object tags of each object;

[0097] Figure 12 It shows Figure 3 A schematic diagram of the object tag list formed by integrating object tags in step 330;

[0098] Figure 13 It shows Figure 3 Step 340 is a flowchart of selecting target objects from the object set based on the matching degree between target keywords and object tags;

[0099] Figure 14 It shows Figure 13 The flowchart for step 1310, determining the matching degree between target keywords and object tags;

[0100] Figure 15 It shows Figure 13 The flowchart for step 1320, selecting the target object from the object set;

[0101] Figure 16A -C shows Figure 13 A schematic diagram illustrating an implementation process for step 1320, selecting a target object from the object set;

[0102] Figure 17 It shows Figure 3 Step 340 is another flowchart for selecting target objects from the object set based on the matching degree between target keywords and object tags;

[0103] Figure 18 It shows Figure 17 Step 1730 is a flowchart of the process of performing a second screening on the objects after the first screening to obtain the target object;

[0104] Figure 19A -B shows Figure 17 A schematic diagram illustrating the implementation process of step 1730, which involves performing a second screening on the objects after the first screening to obtain the target object;

[0105] Figure 20 It shows Figure 17 Another flowchart for step 1730, which involves performing a second screening on the objects after the first screening to obtain the target objects;

[0106] Figure 21A -C shows Figure 17 Step 1730 illustrates another implementation process of performing a second screening on the objects after the first screening to obtain the target object;

[0107] Figure 22 This is a flowchart of a content delivery method according to another embodiment of the present disclosure;

[0108] Figure 23 This is a flowchart of a content delivery method according to another embodiment of the present disclosure;

[0109] Figure 24 This is a flowchart of a content delivery method according to another embodiment of the present disclosure;

[0110] Figure 25 It shows Figure 24 The diagram illustrates the implementation process of the ratios corresponding to each attribute information.

[0111] Figure 26 This is a schematic diagram illustrating the specific implementation process of a content delivery method according to an embodiment of this disclosure;

[0112] Figure 27 This is a schematic diagram illustrating the interaction process between an object and an object terminal in a content delivery method according to an embodiment of this disclosure;

[0113] Figure 28 This is a block diagram of a content delivery device according to an embodiment of the present disclosure;

[0114] Figure 29 This is a terminal structure diagram of a content delivery method according to an embodiment of the present disclosure;

[0115] Figure 30 This is a server structure diagram of a content delivery method according to an embodiment of the present disclosure. Detailed Implementation

[0116] To make the objectives, technical solutions, and advantages of this disclosure clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and are not intended to limit the scope of this disclosure.

[0117] Before providing a further detailed description of the embodiments of this disclosure, the terms and concepts used in these embodiments are explained, and they are subject to the following interpretations:

[0118] Artificial intelligence (AI) is the theory, methods, technology, and application systems that use digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to achieve desired results. In other words, AI is a comprehensive technology within computer science that attempts to understand the essence of intelligence and produce a new kind of intelligent machine that can react in a way similar to human intelligence. AI studies the design principles and implementation methods of various intelligent machines, enabling them to have perception, reasoning, and decision-making capabilities. AI technology is a comprehensive discipline involving a wide range of fields, encompassing both hardware and software technologies. Fundamental AI technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing, operating / interactive systems, and mechatronics. AI software technologies mainly include computer vision, speech processing, natural language processing, and machine learning / deep learning. With the research and advancement of artificial intelligence (AI) technology, AI is being studied and applied in various fields, such as smart homes, smart wearable devices, virtual assistants, smart speakers, smart marketing, autonomous driving, drones, robots, smart healthcare, and smart customer service. It is believed that with the development of technology, AI will be applied in more fields and play an increasingly important role.

[0119] Knowledge graphs are a modern theory that integrates theories and methods from applied mathematics, computer graphics, information visualization, and information science with bibliometric citation analysis and co-occurrence analysis. They utilize visual graphs to illustrate the core structure, development history, cutting-edge fields, and overall knowledge architecture of a discipline, achieving multidisciplinary fusion. The primary goal of knowledge graphs is to describe various entities and concepts existing in the real world, as well as the strong relationships between them. Relationships are used to describe the connections between two entities. From an artificial intelligence perspective, knowledge graphs are tools that use knowledge bases to assist in understanding human language. From a database perspective, knowledge graphs are methods for storing knowledge using graphs. Knowledge graphs are a relatively general formal description framework for semantic knowledge, using nodes to represent semantic symbols and edges to represent the relationships between semantics. Knowledge graphs aim to describe various entities or concepts existing in the real world and their relationships, forming a huge semantic network graph where nodes represent entities or concepts, and edges consist of attributes or relationships. Currently, the term "knowledge graph" is used broadly to refer to various large-scale knowledge bases. Knowledge graphs are also known as semantic networks. From their early days, semantic networks have driven graph-based knowledge representation. For example, in the process of promoting the RDF standard, in such a graph-based knowledge representation system, entities are nodes of the graph, and the connections between nodes are relations.

[0120] An entity is a distinguishable and independently existing thing, such as a person, a city, a plant, or a commodity. Everything in the world is composed of concrete things, which are referred to here as entities. Entities are the most basic elements in a knowledge graph, and different entities have different relationships.

[0121] Relationship: A relationship that exists between entities. A relation is formalized as a function that maps k nodes to a Boolean value. In a knowledge graph, a relation is a function that maps k graph nodes (entities, semantic classes, attribute values) to Boolean values.

[0122] The internet involves a large amount of content push and delivery. Currently, content push and delivery generally adopts a method of finding popular content that matches the browsing history or interests of the target audience, i.e., "people finding things." This is extremely unfriendly to some new content creators or new types of content from established authors. Because they lack the accumulation of historical delivery data, the traditional "people finding things" method will allocate very little exposure to the content, which is detrimental to the efficiency of reaching the target audience for specific content. This forces some content to reach the target audience through additional channels, wasting network resources and reducing the utilization rate of network resources.

[0123] System architecture and scenario description of the embodiments disclosed herein

[0124] Figure 1 This is a system architecture diagram of the content delivery method applied according to an embodiment of the present disclosure. It includes an object terminal 140, an Internet 130, a gateway 120, a content delivery platform server 110, and an object tag library 150 of the content delivery platform server 110, etc.

[0125] The target terminal 140 includes various forms such as desktop computers, laptops, PDAs (personal digital assistants), mobile phones, in-vehicle terminals, home theater terminals, and dedicated terminals. Furthermore, it can be a single device or a collection of multiple devices. The target terminal 140 can communicate with the Internet 130 via wired or wireless means to exchange data. The target terminal 140 includes a content delivery and recommendation system 141.

[0126] When object A operates on the front-end interface of object terminal 140, object terminal 140 will input the content to be delivered uploaded by object A into content delivery recommendation system 141. The content delivery recommendation system 141 will match the content to be delivered with the object tags extracted from the object tag library 150 of content delivery platform server 110, and select several objects from content delivery platform server 110 to deliver the content to be delivered based on the matching results. Finally, the content to be delivered will be delivered to the selected objects.

[0127] Content delivery platform server 110 refers to a computer system that can provide certain services to target terminal 140. Compared with ordinary target terminal 140, content delivery platform server 110 has higher requirements in terms of stability, security, and performance. Content delivery platform server 110 can be a single high-performance computer in a network platform, a cluster of multiple high-performance computers, a portion of a single high-performance computer (e.g., a virtual machine), or a combination of portions of multiple high-performance computers (e.g., virtual machines).

[0128] Gateway 120, also known as an internetwork connector or protocol converter, is a computer system or device that acts as a translator, enabling network interconnection at the transport layer. It bridges the gap between two systems using different communication protocols, data formats, languages, or even completely different architectures. Gateways can also provide filtering and security functions. Messages sent from target terminal 140 to content delivery platform server 110 are forwarded to the corresponding content delivery platform server 110 via gateway 120. Messages sent from content delivery platform server 110 to target terminal 140 are also forwarded to the corresponding target terminal 140 via gateway 120.

[0129] The embodiments disclosed herein can be applied in various scenarios, such as Figure 2A -K indicates video delivery scenarios, etc.

[0130] like Figure 2A As shown, Subject A opens a social media application on Subject Terminal 140 and enters the personal center interface within the application. The personal center interface displays browsing settings and information such as "My Video Account." "My Video Account" includes Subject A's nickname, profile picture, number of followers, and video account messages. At the bottom of the interface are displayed "Publish Video" and "Start Live Stream" buttons. When Subject A clicks the "Publish Video" button, the video publishing process begins; when Subject A clicks the "Start Live Stream" button, the live streaming process begins. At this point, Subject A clicks the "Publish Video" button again to begin the video publishing process.

[0131] like Figure 2B As shown, when object A clicks the "Publish Video" button, the interface will display a "Shoot" button, a "Select from Album" button, and a "Cancel" button. The "Shoot" button allows the object to shoot a video in real-time; the "Select from Album" button allows the object to select an existing video from its album for publication. At this point, object A clicks the "Select from Album" button to enter the video selection process.

[0132] like Figure 2C As shown, when object A clicks the "Select from album" button, the interface will display multiple pictures and videos contained in the album. Object A selects the first content in the album and clicks the "Next" button to confirm the selection of the first content for publication.

[0133] like Figure 2D As shown, when object A selects the first content and clicks the "Next" button, the interface will display the first content, an input box for adding text descriptions, and selection controls such as "Location," "Activity," and "Link." At this point, object A enters a text description of the first content in the input box and clicks the "Discover" button to enter the target audience filtering process.

[0134] like Figure 2EAs shown, when object A clicks the "Mining" button, the target audience selection process begins. During this process, the content recommendation system 141 in object terminal 140 receives the first content, extracts keywords from it, and displays a keyword list on the interface. The displayed keyword list contains six target keywords: Keyword A, Keyword B, Keyword C, Keyword D, Keyword E, and Keyword F. Keyword A has a weight of 0.05, Keyword B has a weight of 0.15, Keyword C has a weight of 0.45, Keyword D has a weight of 0.1, Keyword E has a weight of 0.2, and Keyword F has a weight of 0.05. Object A can click the "Modify" button to adjust the target keywords and their weights in the keyword list, or click the "OK" button to proceed directly to the next step. At this point, object A can click the "Modify" button to adjust the target keywords and their weights in the keyword list.

[0135] like Figure 2F As shown, after object A clicks the "Modify" button, object A can adjust the target keywords and their weights on the interface. First, object A adjusts the target keywords in the keyword list to keyword Q, keyword B, keyword C, keyword D, keyword E, and keyword K. Next, object A adjusts the weight of keyword Q to 0.2, keyword B to 0.05, keyword C to 0.4, keyword D to 0.05, and keyword K to 0.1. After the adjustments are complete, object A clicks the "OK" button to enter the audience filtering stage.

[0136] like Figure 2G As shown, during the audience screening stage, the interface of the target terminal 140 displays various screening criteria and corresponding input boxes for the target audience. These criteria include audience activity level, education level, and geographic location. Target A enters "≥Medium Activity Level" in the "Audience Screening" input box, "≥Junior High School" in the "Education Level" input box, and "City A" in the "Geographic Location" input box. At this point, the target audience consists of individuals in City A with a junior high school education or higher and an activity level no lower than medium activity level. Target A clicks the "OK" button to confirm the audience screening criteria.

[0137] like Figure 2HAs shown, when object A clicks the "OK" button, the content delivery recommendation system 141 of object terminal 140 will filter each object in the content delivery platform server 110 according to the received target audience filtering conditions, obtain the target audience that meets the conditions, and generate an overall distribution map based on the target audience's activity level, education level, and other attribute information. Then, the distribution of the target audience that meets the conditions will be displayed to object A. Specifically, there are a total of 80 people who meet the conditions. The overall distribution includes the activity level percentage and the education level percentage. In terms of activity level, there are 50 highly active people and 30 moderately active people. In terms of education level, there are 10 people with junior high school education, 40 people with university education, and 30 people with high school education. At this time, object A can click the "Previous" button to return to the filtering condition setting stage and reset the filtering conditions; object A can also click the "OK" button to confirm that the 80 people who meet the conditions are the final target audience.

[0138] like Figure 2I As shown, when Object A clicks the "OK" button, the content publishing process begins. During this process, the interface displays the first content, a text description of the first content entered by Object A, and selection controls for "Location," "Activity," and "Link." At this point, Object A clicks the "Publish" button to confirm the publication of the first content.

[0139] like Figure 2J As shown, when object A clicks the "Publish" button, a pop-up window will appear on the interface. The pop-up window displays the prompt field "Do you want to send the first content to the 60 eligible objects?" At this time, object A clicks the "Yes" button, and object terminal 140 responds to object A's selection and sends the first content.

[0140] like Figure 2K As shown, after the first piece of content is published, a pop-up window will appear on the interface, displaying the message "Content has been published." At this point, object A clicks the "OK" button to confirm the publication of the first piece of content.

[0141] General Description of Embodiments in this Disclosure

[0142] According to one embodiment of this disclosure, a content delivery method is provided.

[0143] This content delivery method can be used for Figure 2AThe video delivery scenario shown in -K can also be used for audio content recommendation, information recommendation, and other scenarios. Currently, this disclosure provides a scheme that determines target keywords based on the content to be delivered, and selects matching target objects for the content based on the matching degree between the target keywords and the object tags of various objects in the platform object set. This can quickly reach potentially interested target objects for the content to be delivered, improving the efficiency of reaching specific content and the accuracy of content delivery.

[0144] like Figure 3 As shown, a content delivery method according to an embodiment of this disclosure may include:

[0145] Step 310: Based on the number of published content of the same type as the first content to be delivered and the number of published content of the first object to which the first content is delivered, determine that the first condition is met;

[0146] Step 320: In response to the first condition being met, determine the first keyword based on the first content, expand the second keyword based on the first keyword, and integrate the first keyword and the second keyword into the target keyword;

[0147] Step 330: Obtain the object tags of each object in the object set on the content delivery platform;

[0148] Step 340: Select target objects from the object set based on the matching degree between target keywords and object tags;

[0149] Step 350: Deliver the first content based on the selected target audience.

[0150] Steps 310-350 are briefly described below.

[0151] In step 310, the first object refers to an object that wants to share certain content with other objects via the Internet. The first content refers to the video, image, text, etc., that the first content wants to publish.

[0152] The number of published content refers to the total number of times the first object has published videos, images, and text on the content delivery platform.

[0153] The first condition refers to the requirements that should be followed when delivering content based on keyword matching.

[0154] In step 320, the first keyword is a core information word that can describe the first content. The second keyword is a word that is semantically similar to the first keyword.

[0155] In step 330, the content delivery platform refers to a platform that enables individuals to share the content they want to publish with other individuals. For example, commonly used content delivery platforms include video accounts, official accounts, social media applications, etc.

[0156] Object tags refer to standardized definitions of object attributes, behaviors, preferences, etc., used to describe the object characteristics of various objects on a content delivery platform.

[0157] In step 340, the matching degree is a feature that can describe the degree of similarity between the target keyword and the object label. The matching degree can be expressed in numerical, vector or other forms.

[0158] The target audience is the people on the content delivery platform who may be interested in the first piece of content.

[0159] In step 350, the process of delivering the first content based on the selected target object includes, but is not limited to, the following steps:

[0160] Establish communication connections with content delivery platforms;

[0161] The first content is delivered to the target audience of the content delivery platform.

[0162] In this specific implementation, the target terminal first establishes a communication connection with the content delivery platform. Then, according to a preset frequency and preset time, it delivers first content to the target object on the content delivery platform, exposing the first content to the target object so that the target object can perform operations such as browsing, clicking, downloading, forwarding, or liking the first content. The preset frequency and preset time can be set according to actual needs and are not limited.

[0163] Furthermore, when delivering the first content to the target audience, it can be displayed as a pop-up notification, or it can be displayed in the center of the browsing page so that the target audience notices the first content first. Other methods can also be used to display the first content without restriction.

[0164] Through steps 310-350 above, in this embodiment of the disclosure, when the first condition is determined to be met based on the number of published content of the same type as the first content to be delivered and the number of published content of the first object to which the first content is delivered, it indicates that the author of the first content is relatively new, or the type of the first content is relatively new. At this time, due to the lack of historical data accumulation, the first content will be allocated less exposure using the traditional model, reducing the content reach efficiency. This embodiment of the disclosure determines the first keyword based on the first content and expands it into a second keyword. On this basis, it integrates them into a target keyword, and selects the matching target object for the first content based on the matching degree of the target keyword with the object tags of each object in the platform object set. This delivery method does not rely on the accumulation of historical data of the first object, and can quickly reach potentially interested target objects for the first content, improving the object reach efficiency of specific content, without having to reach through other means, thus improving the utilization rate of network resources. The automatic generation of keywords improves the automation of content delivery, and the joint determination of the target object based on the basic first keyword and the expanded second keyword improves the accuracy of content delivery.

[0165] The above is a general description of steps 310-350. Since step 350 has been detailed in the above general description, the specific implementations of steps 310, 320, 330 and 340 will be described in detail below.

[0166] Detailed description of step 310

[0167] In step 310, the first condition is determined to be met based on the number of published content of the same type as the first content to be delivered and the number of published content of the first object delivering the first content.

[0168] In this specific implementation, when the first object has a relatively low number of publications, it indicates that the first object is a relatively new author. If the first content is directly submitted to the content delivery platform, the likelihood of the first content published by the first object reaching interested target audiences is low. Therefore, the number of publications by the first object can be used to determine whether the first condition is met. When the first condition is met, the content delivery method of this embodiment is used to increase the exposure of the content published by the first object, thereby enabling the content published by the first object to reach potentially interested target audiences.

[0169] On the other hand, a high number of posts by the first object indicates that the first object is an author with a certain level of exposure. If the first object posts newer content, the limited number of previously published content of the same type will affect the exposure of the first content, as it will affect the identification of interested target audiences. Therefore, the number of previously published content of the same type as the first content can be used to determine whether the first condition is met. When the first condition is met, the content delivery method of this embodiment is used to increase the exposure of the newer content, enabling it to reach potentially interested target audiences.

[0170] It should be noted that, in this embodiment of the disclosure, when the number of published content of the same type as the first content to be published and the number of published content of the first object to publish the first content are obtained, the individual permission or individual consent of the first object will be obtained through pop-up windows or redirection to a confirmation page, etc. After the individual permission or individual consent of the first object is clearly obtained, the number of published content of the same type as the first content to be published and the number of published content of the first object to publish the first content are obtained.

[0171] like Figure 4 As shown, in some embodiments, the process of determining that the first condition is met in step 310 includes, but is not limited to, the following steps 410 to 440.

[0172] Step 410: If the number of published contents of the first object that delivers the first content is less than the first threshold, then the first condition is determined to be met.

[0173] Step 420: Determine the first type of the first content;

[0174] Step 430: Determine the number of published content of the first type on the content delivery platform;

[0175] Step 440: If the number of published content of the first type is less than the second threshold, then the first condition is determined to be met.

[0176] Steps 410-440 are described in detail below.

[0177] In step 410, a threshold comparison can be used to determine whether the first object is a relatively new author. Specifically, the number of published content by the first object that posted the first content is obtained from the object's terminal log. The object's terminal log records data generated by the first object's past publishing, browsing, forwarding, liking, and other object behaviors. Next, the number of published content by the first object is compared with a first threshold to obtain a first comparison result. Finally, if the first comparison result shows that the number of published content by the first object is less than the first threshold, it indicates that the first object is a relatively new author, and the exposure of the content published by the first object needs to be increased. Therefore, the first condition is determined to be met. If the first comparison result shows that the number of published content by the first object is not less than the first threshold, it indicates that the first object is not a relatively new author, but an author with a certain amount of exposure. In order to make the first content published by the first object trigger interested target objects as much as possible, it is necessary to further determine whether the first condition is met based on the type of the first content.

[0178] In step 420, the first object can select a type from the preset types displayed on the front-end interface of the object terminal as the first type of the first content. The object terminal can respond to the first object's selection operation and determine the first type of the first content. Specifically, the preset types include multi-level tags, and each first-level tag has second-level tags. For example, the first-level tags include music, food, travel, film and television, sports, etc. When the first-level tag is music, the subordinate second-level tags include pop music, rock music, folk music, etc.; when the first-level tag is food, the subordinate second-level tags include Chinese food, Western food, etc. When the first object selects "food - Chinese food" from the multiple preset types displayed on the front-end interface of the object terminal, the object terminal can determine the first type of the first content as Chinese food.

[0179] In step 430, the target terminal may send a data retrieval request to the content delivery platform to request the number of published content of the first type on the content delivery platform. Specifically, the target terminal first sends a data retrieval request to the content delivery platform. Based on the data retrieval request, the content delivery platform can query the backend server to find all published content of the first type, count the number of published content of the first type, and send a response containing the number of published content of the first type to the target terminal, thereby enabling the target terminal to determine the number of published content of the first type on the content delivery platform.

[0180] It should be noted that, in this embodiment of the disclosure, when the target terminal sends a data acquisition request to the content delivery platform, it obtains the first target's individual permission or consent through pop-ups or redirection to a confirmation page. Only after obtaining the first target's individual permission or consent does it send the data acquisition request to the content delivery platform. Similarly, when the content delivery platform provides feedback on the number of published content of the first type based on the data acquisition request, it also obtains the first target's individual permission or consent through pop-ups or redirection to a confirmation page. Only after obtaining the first target's individual permission or consent does it provide the target terminal with the number of published content of the first type.

[0181] In step 440, after determining the number of published content of the first type, the target terminal compares the number of published content of the first type with a second threshold to obtain a second comparison result. If the second comparison result shows that the number of published content of the first type is less than the second threshold, it indicates that the first type to which the first content belongs is a relatively new type, and the exposure of the first content needs to be increased. Therefore, the first condition is determined to be met. If the second comparison result shows that the number of published content of the first type is not less than the second threshold, it indicates that the first type to which the first content belongs is a type familiar to the objects on the content delivery platform. The content delivery platform has accumulated a certain number of target objects interested in the first type, and the probability that the first content can reach the target objects of interest is high. Therefore, it is not necessary to use the content delivery method of this embodiment to increase the exposure of the first content. Therefore, the first condition is determined to be unmet.

[0182] It should be noted that the first and second thresholds mentioned above can be set according to the actual situation, without specific restrictions.

[0183] Through the above steps 410-420, this embodiment of the disclosure can determine whether the first object to which the first content is being delivered is a relatively new author by using a threshold comparison method based on the number of published content of the first object. And when the first object is not a relatively new author, it can determine whether the first content is a relatively new type of published content by using a threshold comparison method based on the number of published content of the first type on the content delivery platform. This method can effectively identify the first content published by a relatively new author or the first content of a relatively new type, which is conducive to quickly reaching potentially interested target objects for the first content and improving the object reach efficiency of the first content.

[0184] In one specific embodiment, the process of determining that the first condition is met in step 310 includes, but is not limited to, the following steps:

[0185] Display the trigger control;

[0186] In response to the selection of the trigger control, the first condition is determined to be met based on the number of published content of the same type as the first content to be delivered and the number of published content of the first object delivering the first content.

[0187] The trigger control is a visual control on the front-end interface of the object terminal. The trigger control is used to provide the first object with a choice. When the first object selects the trigger control, it will enter the process of determining whether the first condition is true.

[0188] In this specific implementation, when the first object triggers the content delivery process, the object terminal will display a trigger control on the interface for the first object. If the first object selects the trigger control, the object terminal will respond to the first object's selection operation and determine whether the first condition is met based on the number of published content of the same type as the first content and the number of published content of the first object. When the first condition is met, the content delivery method of this embodiment is used to deliver the first content.

[0189] The specific implementation process of this embodiment is similar to steps 410-420 described above. The difference is that in this embodiment, the determination process of the first condition is initiated based on the first object's selection of the trigger control after the first object triggers the content delivery process, while the determination process of the first condition in steps 410-420 described above is initiated directly after the first object triggers the content delivery process. To save space, it will not be described in detail again.

[0190] This approach can determine whether the first condition is met based on the first object's selection of the trigger control, allowing the first object to autonomously choose whether to initiate the judgment process of the first condition, thereby improving the personalization of the content delivery method in this embodiment.

[0191] Detailed description of step 320

[0192] In step 320, in response to the first condition being met, a first keyword is determined based on the first content, a second keyword is expanded based on the first keyword, and the first keyword and the second keyword are integrated into a target keyword.

[0193] In the specific implementation of this embodiment, if the first condition is met, it indicates that the first content was published by a relatively new author or is of a relatively new type. If the first content is directly delivered to objects on the content delivery platform, the likelihood of the first content reaching interested target objects is low, which is not conducive to increasing the exposure of the first content. Based on this, this embodiment of the disclosure considers a scheme to determine the target objects of the first content based on keyword matching, so as to improve the matching degree between the first content and the target objects, thereby improving the accuracy of content delivery.

[0194] In practical applications, when publishing content, users often input a text description of the content. This description provides a brief overview of the content and covers its key and important information. Therefore, this embodiment of the disclosure considers a scheme to automatically generate keywords for published content based on the text description, which improves the automation of keyword extraction, thereby enhancing its convenience and accuracy. Furthermore, this embodiment also considers a scheme to assign a corresponding weight to each keyword, which improves the computational efficiency of the keyword matching process, thereby increasing content delivery efficiency.

[0195] The following is combined with Figure 5 and Figure 6 The process of determining the first keyword based on the first content is described in detail.

[0196] In some embodiments, the process of determining the first keyword based on the first content includes, but is not limited to, the following steps 510 to 540.

[0197] Step 510: Extract the first sub-keyword from the text description of the first content;

[0198] Step 520: Extract the second sub-keyword from the text description of the second content placed before the first content by the first object;

[0199] Step 530: Based on the multimodal parsing of the first content, extract the third sub-keyword;

[0200] Step 540: Combine the first sub-keyword, the second sub-keyword, and the third sub-keyword into the first keyword.

[0201] Steps 510-540 are described in detail below.

[0202] In step 510, the text description of the first content entered by the first object on the object's terminal interface is first obtained. Then, machine learning can be used to extract the first sub-keyword from the text description of the first content. The process of extracting the first sub-keyword from the text description of the first content using machine learning includes, but is not limited to, the following steps:

[0203] The text description of the first content is processed by word segmentation and part-of-speech tagging to obtain candidate keywords;

[0204] A pre-trained BERT module is used to semantically encode candidate keywords, generating a candidate keyword vector corresponding to each candidate keyword.

[0205] For each candidate keyword, based on the candidate keyword vector, calculate the relevance between the candidate keyword and the first type of the first content;

[0206] Candidate keywords are ranked according to their relevance, and the candidate keywords with the highest predetermined ranking are selected as the first sub-keywords.

[0207] In this specific implementation, the text description of the first content is first segmented into words according to conventional grammatical rules, breaking it down into characters, words, or phrases. Next, each character, word, or phrase is tagged with its part of speech. Then, meaningless characters in the segmented data are deleted; meaningless characters refer to punctuation marks generated during word segmentation. Stop words are removed from the segmented data based on the stop word list provided by the content delivery platform. Finally, based on the part of speech of the segmented data, only nouns, adjectives, and verbs with a length of at least two are retained, and these retained nouns, adjectives, and verbs with a length of at least two are selected as candidate keywords.

[0208] Furthermore, a pre-trained BERT module is used to semantically encode the candidate keywords, extract the semantic information of each candidate keyword, and obtain the candidate keyword vector corresponding to each candidate keyword.

[0209] It should be noted that the pre-trained BERT module (Bidirectional Encoder Representations from Transformers, BERT) is a commonly used module in machine learning. The pre-trained BERT module in this embodiment is implemented based on transformers and is a bidirectional encoding model. When using the pre-trained BERT module to semantically encode candidate keywords, the candidate keywords are first encoded forward from left to right to obtain a forward encoded feature vector; then, the candidate keywords are encoded backward from right to left to obtain a backward encoded feature vector; finally, the forward encoded feature vector and the backward encoded feature vector are fused to obtain the candidate keyword vector corresponding to the candidate keyword pair.

[0210] Furthermore, a pre-defined Softmax function can be used to calculate the probability distribution vector of each candidate keyword vector in the first type. This probability distribution vector is then used as the relevance between the candidate keyword and the first type of the first content. The larger the probability distribution vector, the higher the relevance of the candidate keyword to the first type.

[0211] It should be noted that the Softmax function is a normalization exponential function that can "compress" a K-dimensional vector z containing arbitrary real numbers into another K-dimensional real vector σ(z), such that each element is in the range (0,1) and the sum of all elements is 1. This function is often used in multi-class classification problems.

[0212] Finally, the candidate keywords are sorted in descending order of relevance, with more relevant keywords appearing first and less relevant keywords appearing last. Based on this order, the candidate keywords with the highest predetermined ranking are selected as the first sub-keywords. The predetermined ranking is set based on actual circumstances and is not limited; for example, a predetermined ranking of 5 could be used, selecting the top five most relevant candidate keywords as the first sub-keywords.

[0213] In step 520, since the first object often accumulates a certain amount of content over time, the second content delivered by the first object before the first content can, to some extent, reflect the content delivery preferences of the first object. Therefore, the second content delivered by the first object before the first content can also be used as a basis for determining keywords. Based on this, the text description of the second content delivered by the first object before the first content is first obtained from the object's terminal log. Then, based on the text description of the second content, the second sub-keyword is extracted.

[0214] Specifically, the process of extracting the second sub-keyword based on the textual description of the second content in step 520 is similar to the process of extracting the first sub-keyword based on the textual description of the first content in step 510. The difference lies in that step 520 extracts the second sub-keyword based on the textual description of the second content, while step 510 extracts the first sub-keyword based on the textual description of the first content; the textual descriptions used for keyword extraction differ between the two. For the sake of brevity, this will not be elaborated further here.

[0215] In step 530, when the first content is in video format, it often contains multiple modalities such as video, audio, and image, and the keywords of the first content under each modality may differ. Directly extracting keywords from a single modality often leads to inaccurate keywords. Therefore, this embodiment of the present disclosure considers a scheme to extract keywords by performing multimodal parsing on the first content. Specifically, the specific process of extracting the third sub-keyword based on multimodal parsing of the first content in this embodiment of the present disclosure includes, but is not limited to, the following steps:

[0216] The first content is processed by extracting video frames to obtain video frames, and visual features are obtained by extracting features based on the video frames.

[0217] The first content is resampled to obtain preliminary audio, and features are extracted based on the preliminary audio to obtain audio features;

[0218] Keyword extraction was performed based on visual and audio features to obtain the third set of keywords.

[0219] In this specific implementation, the first content is first divided into several video segments of equal duration. Then, a predetermined number of video frames are extracted from each video segment to obtain video frames for the first content. Next, a pre-defined convolutional neural network is used to extract the two-dimensional and three-dimensional visual features of the video frames, and these features are fused to obtain the final visual features. Further, the first content is resampled to obtain preliminary audio, which is mono audio. Then, a short-time Fourier transform is performed on the preliminary audio to obtain a spectrogram; a pre-defined Mel filter bank is used to calculate the spectrum of the spectrogram to obtain a Mel spectrum; the logarithm of the Mel spectrum is taken and the frames are grouped to obtain the final audio sequence. Next, a pre-defined audio extraction network is used to extract features from the audio sequence to obtain audio features. Finally, based on a pre-defined visual feature lookup table and visual features, visual keywords are extracted; based on a pre-defined audio feature lookup table and audio features, audio keywords are extracted; and the audio keywords and visual keywords are integrated into a third sub-keyword.

[0220] Audio features can be composed of phonemes, which are the smallest units of speech defined based on the natural attributes of speech. They are analyzed based on the articulation actions within a syllable, with each action constituting a phoneme. Visual features can include various feature information such as color, texture, text, and motion information, as mentioned in the first part.

[0221] It should be noted that the pre-defined convolutional neural network is constructed based on a ResNet-152 convolutional neural network pre-trained on the ImageNet dataset and a 3D-ResNet convolutional neural network pre-trained on the Kinetics dataset. The convolutional neural network includes two-dimensional and three-dimensional convolutional parts. The two-dimensional convolutional part is composed of the pre-trained ResNet-152 convolutional neural network, and the three-dimensional convolutional part is composed of the pre-trained 3D-ResNet convolutional neural network. The two-dimensional convolutional part of the convolutional neural network extracts two-dimensional visual features of the video frames, and the three-dimensional convolutional part extracts three-dimensional visual features of the video frames.

[0222] It should be noted that the preset audio extraction network is a VGGish network pre-trained on AudioSet. A preset visual feature lookup table is used to indicate the correspondence between visual features and candidate keywords, and a preset audio feature lookup table is used to indicate the correspondence between audio features and candidate keywords.

[0223] In a specific implementation scenario, the first content is first divided into several 10-second video segments. Eight video frames are then extracted from each segment, and all frames are integrated to obtain a video frame set. Next, the 2D convolutional part of a convolutional neural network (CNN) is used to extract the 2D visual features of the video frames, and the 3D convolutional part is used to extract the 3D visual features. The 2D and 3D visual features are then added together to obtain the visual features. Further, the first content is resampled to obtain preliminary audio at a frequency of 16kHz. Then, a short-time Fourier transform is performed on the preliminary audio using a 25ms Hann window and a 10ms frame shift to obtain a spectrogram. A 64th-order Mel filter bank is used to calculate the acoustic spectrum of the spectrogram, obtaining the Mel acoustic spectrum. Finally, the logarithm of the Mel acoustic spectrum is taken with a bias of 0.01 to obtain a stable Mel acoustic spectrum. The stable Mel acoustic spectrum is then framed for a duration of 0.96 seconds to obtain the final audio sequence. In this audio sequence, each frame contains 64 Mel bands (excluding frame overlap). Next, a pre-defined audio extraction network is used to extract features from the audio sequence, yielding audio features. Finally, candidate keywords corresponding to the visual features are queried from a visual feature lookup table and used as visual keywords; similarly, candidate keywords corresponding to the audio features are queried from an audio feature lookup table and used as audio keywords; the audio keywords and visual keywords are then integrated into a third set of sub-keywords.

[0224] like Figure 6 As shown, when performing multimodal analysis on the first content and extracting the third sub-keywords, the first content in video format is first decomposed into multiple modalities. Since the first content in video format includes both audio and video modalities, it is decomposed into two parts based on the modality: the audio part and the video part. Next, keywords are extracted from the audio part of the first content, resulting in four keywords: Keyword 1, Keyword 2, Keyword 3, and Keyword 4. Keywords are also extracted from the video part of the first content, resulting in four keywords: Keyword 2, Keyword 4, Keyword 5, and Keyword 6. Finally, all keywords are integrated to obtain the third sub-keywords of the first content. There are a total of six third sub-keywords: Keyword 1, Keyword 2, Keyword 3, Keyword 4, Keyword 5, and Keyword 6.

[0225] In step 540, when integrating the first, second, and third sub-keywords into the first keyword, firstly, all the first, second, and third sub-keywords are included in the same set. Next, duplicate keywords in the set are deduplicated, removing multiple identical keywords so that each keyword appears only once, thus obtaining the final first keyword.

[0226] It should be noted that each primary keyword has a primary weight, and the primary weight of the primary keyword is a fixed weight, which is set according to actual needs.

[0227] Through steps 510-540 above, this embodiment of the disclosure combines the first content and the text description of the second content placed before the first content by the first object, and uses machine learning and other methods to extract keywords. This improves the automation level of keyword extraction, effectively increases the number of keywords, and also improves the accuracy and rationality of the first keywords. Furthermore, this embodiment of the disclosure uses a multimodal parsing method to extract keywords for first content with multiple modalities, such as videos, enabling the extraction of keywords from the first content in different modalities, achieving multi-dimensional keyword extraction. Multimodal keyword extraction improves the richness and accuracy of the first keywords.

[0228] When the first target is a relatively new author, relying solely on the first content and the second content submitted by the first target before the first content often results in a limited number of keywords. If the number of keywords is insufficient, adequate coverage may not be achieved during keyword matching, affecting the accuracy of keyword matching. This disclosure provides a scheme for expanding keywords based on the first keyword, which increases the number of keywords, improves keyword richness, and thus enhances the accuracy of keyword matching.

[0229] The following is combined Figure 7 and Figure 8A -B provides a detailed description of the process of expanding the second keyword based on the first keyword.

[0230] In some embodiments, the process of expanding a second keyword based on a first keyword includes, but is not limited to, the following steps 710 to 730.

[0231] Step 710: Input the first keyword into the related word prediction model to obtain the fourth sub-keyword;

[0232] Step 720: Expand the first keyword using a knowledge graph to obtain the fifth sub-keyword;

[0233] Step 730: Combine the fourth and fifth sub-keywords into the second keyword.

[0234] Steps 710-730 are described in detail below.

[0235] In step 710, the related word prediction model can be a model built based on large-scale language models such as GPT-3, T5, and BERT, and this model has autonomous learning capabilities. When training the related word prediction model, several sample words are first input into the model. The model performs semantic analysis and word prediction on the sample words, learns all the knowledge points contained in the sample words, and generates predicted related words for each sample word. Then, the model parameters are adjusted based on the semantic similarity between the predicted related words and the sample words to update the model. After training, the first related word is input into the model for semantic analysis and word prediction, generating related words corresponding to each first related word, and these related words are used as the fourth sub-keywords. Each fourth sub-keyword carries a confidence probability, which reflects the semantic similarity between the fourth sub-keyword and the first keyword.

[0236] like Figure 8A As shown, the first keyword A is first input into the related word prediction model. The model then performs semantic analysis and word prediction on keyword A, obtaining three related words that are semantically close to keyword A. These three related words are a1, a2, and a3. The confidence probability of related word a1 is 1, that of related word a2 is 2, and that of related word a3 is 3. Finally, related words a1, a2, and a3 are used as the fourth sub-keyword.

[0237] In step 720, when expanding the first keyword using a knowledge graph, the position of the first keyword in the knowledge graph is first determined. Then, based on the position, the first path passing through the first keyword is searched in the knowledge graph. The two endpoints of the first path are a word node at the top level and a word node at the bottom level of the knowledge graph, respectively. Finally, the words corresponding to each node on the first path are used as the fifth sub-keyword.

[0238] like Figure 8BAs shown, the knowledge graph is divided into four levels. The first level is entertainment; the second level includes music and sports within the entertainment category; the third level includes Chinese and Western music within the music category, and tennis, basketball, and football within the sports category; the fourth level includes pop and rock within the Chinese music category, and rock, jazz, and pop within the Western music category. Assuming the primary keyword is "Chinese," when expanding on this keyword using the knowledge graph, the first step is to determine its position within the graph. Next, starting from the primary keyword "Chinese," a search is performed upwards or downwards to determine the first path passing through it. This first path includes "Pop-Chinese-Music-Entertainment" and "Rock-Chinese-Music-Entertainment." Finally, the words along the first path are used as the fifth sub-keywords, resulting in "Chinese-Pop," "Chinese-Rock," "Music," and "Entertainment."

[0239] In step 730, the specific process of integrating the fourth and fifth sub-keywords into the second keyword is similar to the specific implementation process of integrating the first, second, and third sub-keywords into the first keyword in step 540 above. To save space, it will not be described in detail here.

[0240] In some embodiments, after integrating the fourth and fifth sub-keywords into a second keyword, the content delivery method of this disclosure further includes determining a first weight for the second keyword, specifically including but not limited to the following steps:

[0241] The confidence probability of the fourth sub-keyword output by the related word prediction model is used as the first weight of the fourth sub-keyword;

[0242] The first weight of the fifth sub-keyword is determined based on the distance between the fifth sub-keyword and the first keyword in the knowledge graph;

[0243] The first weight of the second keyword is determined based on the first weight of the fourth and fifth sub-keywords.

[0244] In this specific implementation, after the first keyword is input into the related word prediction model, the fourth sub-keyword output by the model carries a confidence probability. This confidence probability reflects the semantic similarity between the fourth sub-keyword and the first keyword. Therefore, the confidence probability of the fourth sub-keyword output by the related word prediction model is directly used as the first weight of the fourth sub-keyword. The distance between the fifth sub-keyword and the first keyword reflects their semantic similarity. The closer the fifth sub-keyword is to the first keyword and the fewer the levels they are separated by, the higher their semantic similarity. Conversely, the farther apart they are and the more levels they are separated by, the lower their semantic similarity. Therefore, based on the distance between the fifth sub-keyword and the first keyword, the first weight of the fifth sub-keyword closer to the first keyword is set to a larger value, while the first weight of the fifth sub-keyword farther from the first keyword is set to a smaller value. After determining the first weight of the fourth sub-keyword and the first weight of the fifth sub-keyword, the fourth and fifth sub-keywords are combined into the second keyword, and the first weight of the fourth and fifth sub-keywords is used as the first weight of the obtained second keyword.

[0245] Through steps 710-730 above, this embodiment of the disclosure adopts a combination of knowledge graph and keyword prediction model to achieve keyword expansion, which can effectively increase the number of keywords, improve keyword richness, and thus improve the accuracy of keyword matching.

[0246] It should be noted that although this embodiment illustrates the implementation process of keyword expansion by combining knowledge graphs and keyword prediction models, in practical applications, in order to improve the efficiency of keyword expansion, it is also possible to use knowledge graphs alone or keyword prediction models alone for keyword expansion. The specific implementation process is as described in steps 710 and 720 above. To save space, these will not be repeated.

[0247] In some embodiments, when integrating the first keyword and the second keyword into target keywords, a table template can first be set up. The header of this table template includes the target keywords and the first weight corresponding to each target keyword. Then, the first keyword, the second keyword, and the corresponding first weight are filled into the table template, thereby integrating the first keyword and the second keyword to obtain a keyword list composed of the target keyword and the first weight of each target keyword. For example... Figure 9As shown, the keyword list contains six target keywords: Keyword A (first weight 0.05), Keyword B (first weight 0.15), Keyword C (first weight 0.45), Keyword D (first weight 0.1), Keyword E (first weight 0.2), and Keyword F (first weight 0.05). This method integrates the first and second keywords as target keywords using a template-filling approach. The keyword list clearly represents each target keyword and its first weight, facilitating visual operations on the target keywords by the first user.

[0248] It should be noted that since the first keyword is directly extracted from the text description of the first content and the second content that was placed before the first content, while the second keyword is obtained by expanding on the first keyword, the importance of the first keyword will be higher than that of the second keyword. Therefore, the first weight of the first keyword is often higher than that of the second keyword.

[0249] Since the target keywords generated by the aforementioned automated method often do not meet the actual requirements of the first object, directly using these target keywords in the subsequent content delivery process often leads to inaccurate content delivery. Therefore, this embodiment also considers a scheme for adjusting the target keywords, so that the first object can easily adjust the target keywords, satisfying personalized settings and improving the flexibility and accuracy of the target keywords.

[0250] In some embodiments, such as Figure 10 As shown, after integrating the first keyword and the second keyword as target keywords, the content delivery method of this disclosure embodiment also includes, but is not limited to, the following steps 1010 to 1020.

[0251] Step 1010: Display target keywords;

[0252] Step 1020: Receive the first object's adjustment of the target keywords.

[0253] Steps 1010-1020 are described in detail below.

[0254] In step 1010, after integrating the first keyword and the second keyword into the target keyword, the target terminal displays the target keyword on the front-end interface so that the first object can view the generated target keyword and adjust the target keyword.

[0255] In step 1020, the first object can adjust the target keywords displayed on the front-end interface according to actual needs. This adjustment process includes operations such as modifying, adding, or deleting target keywords. After the first object adjusts the target keywords, the object terminal can receive the adjustment and determine the final target keywords based on the first object's adjustment operation.

[0256] Through the above steps 1010-1020, the embodiments of this disclosure can conveniently display target keywords to the first object, which can then adjust the target keywords according to actual needs. This enables visual editing of target keywords and also satisfies personalized settings for target keywords, improving the flexibility and accuracy of target keywords. Consequently, it can improve the accuracy of matching target objects based on target keywords and enhance the accuracy of content delivery.

[0257] Detailed description of step 330

[0258] In step 330, the object tags of each object in the object set on the content delivery platform are obtained.

[0259] In this specific implementation, since object tags can clearly reflect the interests and preferences of each object regarding content, when selecting target objects for the first content from the content delivery object set, the final target objects can be determined based on the matching degree between object tags and the target keywords of the first content. Generally, object tags for each object are determined based on a series of object behaviors performed by the object on the content delivery platform. The content delivery platform records behavioral data for each object in the background; therefore, in this embodiment, object tags can be obtained based on the content and the behaviors performed by each object on the content. To improve the reusability of object tags and increase content delivery efficiency, after obtaining object tags, the content delivery platform can also store the object tags of each object in a preset object tag library for easy retrieval at any time.

[0260] It should be noted that, in this embodiment of the disclosure, when the object tags of each object in the object set on the content delivery platform are obtained, the individual permission or consent of the first object is obtained through pop-up windows or redirection to a confirmation page, and the object tags of each object in the object set on the content delivery platform are obtained after the individual permission or consent of the first object is clearly obtained.

[0261] The following is combined Figure 11 and Figure 12 The process of obtaining the object tags for each object is described in detail.

[0262] In some embodiments, the process of obtaining object tags for each object in the object set on the content delivery platform includes, but is not limited to, the following steps 1110 to 1140.

[0263] Step 1110: For each object in the object set, obtain the set of content that the object represents that you are interested in;

[0264] Step 1120: Obtain the third keyword of the content in the content collection;

[0265] Step 1130: For each third keyword, determine the number of content items in the content set that contain the third keyword;

[0266] Step 1140: If the number of contents is greater than the fourth threshold, use the third keyword as the object tag of the object.

[0267] Steps 1110-1140 are described in detail below.

[0268] In step 1110, since an object's actions such as liking, saving, and forwarding published content can be used to determine whether the object is interested in the published content on the content delivery platform, when an object saves a piece of content, it indicates that the object is interested in that content. Furthermore, the application logs of the content delivery platform often record object behavior data such as saving, liking, and forwarding published content on the platform over a past period. Based on this, for each object in the object set, the object terminal can send a request to the content delivery platform to obtain content of interest. Then, the content delivery platform will extract the content from the application logs where the object has performed actions such as saving, liking, and forwarding, and include this series of extracted content into a single set, obtaining the set of content that the object indicates is interested in. Further, the content delivery platform will send the set of content that each object indicates is interested in, along with a response, back to the object terminal, thus allowing the object terminal to obtain the set of content that each object indicates is interested in.

[0269] In step 1120, the specific implementation process of obtaining the third keyword of the content in the content set is similar to the process of extracting the first sub-keyword from the text description of the first content in step 510 above. The difference is that step 1120 extracts the corresponding third keyword for each content in the content set, while step 510 extracts the corresponding first sub-keyword for the first content. To save space, it will not be described in detail here.

[0270] In step 1130, for each object, after obtaining the third keyword of each content in the content set of the object, based on the third keyword, traverse each content in the content set, count the total number of content with the third keyword, and obtain the number of content with the third keyword in the content set.

[0271] For example, if object A's content set contains 5 items, namely content 1, content 2, content 3, content 4, and content 5, and the third keyword of content 1 is keyword M, then we iterate through these 5 items and query whether any of them contain keyword M. Further, after the iteration, we find that content 1, content 3, and content 4 contain keyword M, thus determining that the number of items in the content set with the third keyword is 3.

[0272] In step 1140, since the number of contents for each third keyword can reflect the importance of that third keyword in the content of interest to the object identifier, the object tag can be determined based on the number of contents for the third keyword. Based on this, embodiments of this disclosure employ a threshold comparison method to filter third keywords as object tags. Specifically, the number of contents for the third keyword is compared with a fourth threshold. If the number of contents for the third keyword is greater than the fourth threshold, it indicates that the third keyword is a relatively important word in the content of interest to the object identifier; therefore, the third keyword is used as the object tag. If the number of contents for the third keyword is not greater than the fourth threshold, it indicates that the third keyword is a relatively unimportant word in the content of interest to the object identifier; therefore, the third keyword should not be used as the object tag.

[0273] It should be noted that the fourth threshold mentioned above can be set according to actual needs and is not restricted.

[0274] To facilitate matching target keywords and object tags in subsequent keyword matching, each object tag also has a second weight. Since an object may have one or more object tags, and each object tag appears at different frequencies within the content the object represents that it is interested in, randomly setting the weight of object tags or setting each object tag's weight to a fixed value often fails to accurately reflect the importance of each object tag. Therefore, this disclosure also considers a scheme that determines the second weight of each object tag based on the number of contents corresponding to the third keyword, which can more clearly represent the importance of each object tag, thereby improving the matching accuracy between object tags and target keywords.

[0275] In some embodiments, the process of determining the second weight of each object tag based on the number of contents of the third keyword corresponding to each object tag includes, but is not limited to, the following steps:

[0276] For each object tag, the number of times the content of the third keyword corresponding to the object tag appears is taken as the number of times the object tag appears;

[0277] Count the occurrences of all object tags to get the total number of tag occurrences;

[0278] For each object tag, the number of times the object tag appears is divided by the total number of times the tag appears to obtain the second weight of the object tag.

[0279] In the specific implementation of this example, firstly, the number of content entries corresponding to the third keyword of the object tag is obtained, and this number is taken as the occurrence count of that object tag. Next, the occurrence counts of all object tags are added together to obtain the total number of tag occurrences. Finally, the occurrence count of each object tag is divided by the total number of tag occurrences to obtain the second weight of the object tag.

[0280] For example, object A's tags include tag 1, tag 2, and tag 3. Tag 1 corresponds to 4 occurrences of the third keyword 1, tag 2 corresponds to 5 occurrences of the third keyword 2, and tag 3 corresponds to 11 occurrences of the third keyword 3. Based on this, the frequency of tag 1 is determined to be 4, tag 2 to 5, and tag 3 to 11. Further, the frequency of these three tags is added together, resulting in a total of 4 + 5 + 11 = 20 occurrences. Finally, the frequency of each tag is divided by the total frequency of tags to obtain the second weight of each tag. The second weight of tag 1 is 0.25, the second weight of tag 2 is 0.2, and the second weight of tag 3 is 0.55.

[0281] like Figure 12 The diagram shows a list of object tags generated from the object tags of each object in the object set on the content delivery platform. The object tag column contains four objects: Object 1, Object 2, Object 3, and Object 4. Object 1 contains three tags: Tag 1 (second weight 0.2), Tag 2 (second weight 0.5), and Tag 3 (second weight 0.3). Object 2 contains four tags: Tag 1 (second weight 0.2), Tag 3 (second weight 0.5), Tag 4 (second weight 0.15), and Tag 5 (second weight 0.15). Object 3 contains two tags: Tag 4 (second weight 0.7) and Tag 5 (second weight 0.3). Object 4 contains four tags: Tag 11 (second weight 0.22), Tag 13 (second weight 0.28), Tag 10 (second weight 0.45), and Tag 5 (second weight 0.05).

[0282] Through steps 1110-1140 above, this embodiment of the disclosure can conveniently determine the content of interest represented by each object based on the object behavior data of objects in the object set, and automatically extract the third keyword of each content of interest. Furthermore, it filters out the third keywords that can serve as object tags by using a threshold comparison method. In addition, this embodiment of the disclosure also determines the importance of object tags based on the frequency of each object tag's occurrence in the content set, and sets a second weight for each object tag based on its frequency of occurrence in the content set. This method can significantly improve the accuracy and rationality of the obtained object tags.

[0283] Detailed description of step 340

[0284] In step 340, target objects are selected from the object set based on the matching degree between target keywords and object tags.

[0285] In this specific implementation, since the target keyword has a first weight and the object tag has a second weight, and the target keyword and object tag belong to the same vocabulary system, the first and second weights can be used to calculate the matching degree between the target keyword and the object tag. Furthermore, several target objects are selected from the object set by random sampling or by sequential sampling based on the matching degree. This method improves the efficiency and accuracy of matching degree calculation, and also improves the efficiency and accuracy of target object selection.

[0286] like Figure 13 As shown, in some embodiments, the target keyword has a first weight, the object tag has a second weight, and step 340 includes, but is not limited to, steps 1310 to 1320.

[0287] Step 1310: Determine the matching degree between the target keyword and the object tag based on the target keyword, the first weight, the object tag, and the second weight;

[0288] Step 1320: Select the target object from the object set based on the matching degree of each object in the object set.

[0289] Steps 1310-1320 are described in detail below.

[0290] In step 1310, the process of determining the matching degree between the target keyword and the object tag based on the target keyword, the first weight, the object tag, and the second weight includes, but is not limited to, the following steps:

[0291] For each object in the object set, identify a target object tag that matches the target keyword from the object tags of that object.

[0292] For each target object tag, determine the product of the second weight of the target object tag and the first weight of the matching target keyword;

[0293] The matching degree is obtained by summing the products of the labels of each target object.

[0294] In the specific implementation of this embodiment, firstly, for each object in the object set, the object tag of that object is compared with the target keyword to determine whether the object tag and the target keyword are consistent. If the object tag and the target keyword are consistent, the object tag is considered to be a match for the target keyword, and the object tag that matches the target keyword is taken as the target object tag. Next, for each target object tag, the second weight of the target object tag and the first weight of the target keyword are multiplied to obtain the product of the target object tag. Finally, for each object, the products of all target object tags are summed, and the sum is taken as the object's matching degree.

[0295] Specifically, the kth first content item k The first weight of the j-th target keyword in the text is w. j For the i-th object u in the object set i object u i With the j-th target keyword t j The second weight of the matched target object label is p ij Object u i With the kth first content item k Matching degree S(u) i ,item k ) can be represented as shown in formula (1).

[0296]

[0297] In step 1320, based on the matching degree of each object in the object set, several objects can be randomly selected as target objects; or several objects with higher matching degree or several objects with matching degree higher than a certain threshold can be selected as target objects, so that the target objects selected from the object set are more in line with expectations.

[0298] Through steps 1310-1320 above, this embodiment of the present disclosure can conveniently filter out target object tags that match the target keywords from the object tags of each object, and calculate the matching degree between the object and the first content based on the target object tags and the target keywords, thereby improving the accuracy and rationality of the matching degree calculation. Next, target objects are selected from the object set based on the matching degree, which improves the accuracy of target object selection, making the selected target objects more consistent with the delivery requirements of the first content, and thus improving the accuracy of content delivery.

[0299] Since object tags for each object in a content delivery platform's object set can often be obtained through multiple channels, meaning object tags frequently originate from multiple sources, matching target keywords with object tags from only one channel often results in low keyword matching accuracy. This disclosure provides a scheme for keyword matching based on object tags from multiple sources, which improves the accuracy of matching degree calculation.

[0300] like Figure 14 As shown, in some embodiments, the second weight includes a first sub-weight from the first source and a second sub-weight from the second source, and step 1310 includes, but is not limited to, steps 1410 to 1430.

[0301] Step 1410: Based on the target keyword, first weight, object tag, and first sub-weight, determine the first matching degree corresponding to the first source;

[0302] Step 1420: Based on the target keyword, first weight, object tag, and second sub-weight, determine the second matching degree corresponding to the second source;

[0303] Step 1430: Determine the matching degree based on the first matching degree and the second matching degree.

[0304] Steps 1410-1430 are described in detail below.

[0305] In step 1410, the first sub-weight is the weight of the object tag originating from the first source. Specifically, firstly, for each object in the object set, determine the target object tag that matches the target keyword from the object tags of the first source. Next, for each target object tag, multiply the first sub-weight of the target object tag by the first weight of the matching target keyword to obtain the product result. Finally, sum the product results of all target object tags originating from the first source to obtain the first matching degree corresponding to the first source.

[0306] In step 1420, the second sub-weight is the weight of the object tag originating from the second source. Specifically, firstly, for each object in the object set, a target object tag from the object tags of the second source that matches the target keyword is determined. Next, for each target object tag, the second sub-weight of the target object tag is multiplied by the first weight of the matching target keyword to obtain the product result. Finally, the product results of all target object tags originating from the second source are summed to obtain the second matching degree corresponding to the second source.

[0307] The specific implementation process of steps 1410 and 1420 above is similar to that of step 1310 above. To save space, it will not be described again.

[0308] In step 1430, the first matching degree and the second matching degree can be weighted and calculated to obtain the matching degree. Specifically, firstly, a first matching weight corresponding to the first matching degree and a second matching weight corresponding to the second matching degree are set; then, the first matching weight and the first matching degree are multiplied to obtain a first product result; then, the second matching weight and the second matching degree are multiplied to obtain a second product result; finally, the first product result and the second product result are added together to obtain the matching degree.

[0309] It should be noted that the sum of the first matching weight and the second matching weight is 1. The first matching weight and the second matching weight can be equal or unequal, without any restriction.

[0310] In a specific scenario, the first source is a WeChat Official Account, and the second source is a video account. First, tags are extracted from the object profile from the first source, resulting in tag 1 with a weight of 0.7 and tag 2 with a weight of 0.3. Tags are also extracted from the object profile from the second source, resulting in tag 2 with a weight of 0.4, tag 3 with a weight of 0.5, and tag 4 with a weight of 0.1. Next, since the downstream application is a video application, the adjustment weights for the first and second sources are set to 0.3 and 0.7 respectively, based on their relevance to the downstream application. Each tag is then multiplied by its corresponding adjustment weight to obtain its new weight. Specifically, for the first source, the new weight for tag 1 is 0.21, and for tag 2 it is 0.09; for the second source, the new weight for tag 2 is 0.28, for tag 3 it is 0.35, and for tag 4 it is 0.07. Next, the weights of all tags are summed to obtain the total weight, which is 0.21 + 0.09 + 0.28 + 0.35 + 0.07 = 1. Further, the weights of all tags are divided by the total weight to obtain the first sub-weight of each tag originating from the first source and the second sub-weight of each tag originating from the second source. Specifically, the first sub-weight of tag 1 is 0.21, and the second sub-weight is 0; the first sub-weight of tag 2 is 0.09, and the second sub-weight is 0.28; the first sub-weight of tag 3 is 0, and the second sub-weight is 0.35; and the first sub-weight of tag 4 is 0, and the second sub-weight is 0.07.

[0311] Through the above steps 1410-1430, the embodiments of this disclosure can assign different weights to object tags from different sources, thereby matching object tags from multiple channels with target keywords, which can effectively improve the accuracy of matching degree calculation.

[0312] Randomly selecting several candidate objects from the object set as target objects based on matching degree often results in inaccurate selection. This can lead to target objects that are not necessarily interested in the first content, causing a high degree of deviation in the first content reaching interested target objects and affecting the accuracy of the first content delivery. This disclosure proposes a scheme that uses threshold comparison and ranking selection to filter target objects from the object set. This effectively improves the accuracy of target object filtering, enabling the first content to reach more interested target objects, thereby improving the accuracy of the first content delivery.

[0313] The following is combined Figure 15 and Figure 16A -C describes the specific process of selecting target objects from a set of objects using threshold comparison and sorting selection.

[0314] In one embodiment, target objects can be filtered from the object set solely based on threshold comparisons. Specifically, first, candidate objects with a matching degree greater than a third threshold are identified. Then, these candidate objects with a matching degree greater than the third threshold are selected as target objects. Figure 16A As shown, the object set contains six objects, namely object 1, object 2, object 3, object 4, object 5, and object 6. The matching degree of object 1 is 0.88, that of object 2 is 0.96, that of object 3 is 0.73, that of object 4 is 0.42, that of object 5 is 0.2, and that of object 6 is 0.68. Setting the third threshold to 0.5, candidate objects with matching degrees higher than the third threshold are selected as target objects. Therefore, the target objects include object 1, object 2, object 3, and object 6. This embodiment uses threshold comparison to filter target objects from the object set, achieving clear and simple object filtering and improving the efficiency and accuracy of object filtering.

[0315] In another embodiment, target objects can be filtered from the object set solely based on sorting. Specifically, the objects in the object set are first sorted in descending order of matching degree; then, the objects within the top first position in the sorted list are selected as target objects. Figure 16BAs shown, the object set contains six objects, namely object 1, object 2, object 3, object 4, object 5, and object 6. The matching degree of object 1 is 0.88, that of object 2 is 0.96, that of object 3 is 0.73, that of object 4 is 0.42, that of object 5 is 0.2, and that of object 6 is 0.68. These six objects are sorted according to their matching degree from highest to lowest, resulting in the order: object 2, object 1, object 3, object 6, object 4, and object 5. Selecting the top three objects as the target objects means that the target objects include object 2, object 1, and object 3. This embodiment uses sorting to select target objects from the object set, improving the efficiency and rationality of object selection.

[0316] like Figure 15 As shown, in another embodiment, the process of selecting a target object from the object set based on the matching degree of each object in the object set includes, but is not limited to, the following steps 1510 to 1530.

[0317] Step 1510: In the object set, identify candidate objects with a matching degree greater than the third threshold;

[0318] Step 1520: Sort the candidate objects in descending order of matching degree;

[0319] Step 1530: Select the candidate objects in the first ranking of the sort as the target objects.

[0320] Steps 1510-1530 are described in detail below.

[0321] In step 1510, since a higher matching degree for an object indicates a greater likelihood that the object is interested in the first content, it is more suitable as the target audience for the first content. First, a third threshold is set. Then, the matching degree of each object in the object set is compared with the third threshold. Finally, objects with a matching degree greater than the third threshold are selected as candidate objects.

[0322] In step 1520, the candidate objects are sorted in descending order of matching degree based on their matching degree.

[0323] In step 1530, since a higher matching degree results in a higher position for the candidate object, the candidate object within the first rank in the ranking is selected as the target object. The first rank can be set according to actual needs and is not restricted.

[0324] like Figure 16CAs shown, the object set contains 6 objects, namely object 1, object 2, object 3, object 4, object 5, and object 6. The matching degree of object 1 is 0.88, that of object 2 is 0.96, that of object 3 is 0.73, that of object 4 is 0.42, that of object 5 is 0.2, and that of object 6 is 0.68. First, the third threshold is set to 0.4, and candidate objects with matching degrees higher than the third threshold are selected as candidate objects, including object 1, object 2, object 3, object 4, and object 6. Next, these 5 candidate objects are sorted in descending order of matching degree, resulting in the order: object 2, object 1, object 3, object 6, and object 4. The top three objects in the sorted list are selected as target objects, including object 2, object 1, and object 3.

[0325] Through the above steps 1510-1530, this embodiment of the disclosure combines threshold comparison and sorting selection to filter target objects from the object set. Compared with using threshold comparison or sorting selection alone for object filtering, this method can significantly improve the accuracy of target object filtering, enabling the first content to reach more target objects of interest, thereby improving the accuracy of the first content delivery.

[0326] Because relying solely on keyword matching to filter target objects often results in the initial target audience not being able to adjust the selection based on actual needs, it hinders personalized settings for selecting target objects from a set and affects the accuracy of target object selection. This disclosure proposes a scheme that combines keyword matching and audience filtering to filter target objects. This approach enables personalized target object selection, ensuring that the selected target objects better meet the initial target audience's actual needs, and also improves the accuracy and rationality of target object selection.

[0327] In some embodiments, such as Figure 17 As shown, the process of selecting target objects from the object set based on keyword matching and audience screening includes, but is not limited to, the following steps 1710 to 1730.

[0328] Step 1710: Based on the matching degree between the target keywords and object tags, perform the first screening in the object set to obtain the objects after the first screening;

[0329] Step 1720: Obtain the first filtering criteria for the first object;

[0330] Step 1730: Based on the matching of the first filtering conditions and the attribute information of the first filtered objects, perform a second filtering on the first filtered objects to obtain the target object.

[0331] Steps 1710-1730 are described in detail below.

[0332] In step 1710, the process of performing a first screening on the object set based on the matching degree between the target keywords and object tags to obtain the first-screened objects is similar to the specific implementation process of step 1320 above. The difference is that step 1710 performs a first screening on the object set based on the matching degree to obtain the first-screened objects; while step 1320 filters the target objects from the object set based on the matching degree. To save space, it will not be described in detail.

[0333] In step 1720, the editing interface corresponding to the first filter condition is first displayed to the first object, where the first object inputs the specific content of the first filter condition. Then, the object terminal can receive the first filter condition input by the first object. The specific content of the first filter condition includes filter conditions for tags, filter conditions for user activity levels, filter conditions for user geographic locations, filter conditions for user education levels, filter conditions for account registration years, etc., without restrictions.

[0334] In step 1730, the attribute information of the first-filtered object refers to information such as the object's activity level, location, object tags, education level, and account registration period. First, the attribute information of each first-filtered object is obtained from the log data of the content delivery platform. Next, when performing a second filter on the first-filtered objects based on the matching of the attribute information of the first-filtered objects with the first-filter conditions, each attribute of the first-filtered object is compared with the specific content in the first-filter conditions; if all the attribute information of the first-filtered object matches the specific content of the first-filter conditions, then the first-filtered object is selected as the target object.

[0335] For example, the first screening criterion for the first target is "activity level not lower than medium activity level, account registration period greater than 1 year". There are 5 targets after the first screening. The attribute information of these targets includes activity level and account registration period. These 5 targets are: Target 1 (high activity level, 8 months registration period); Target 2 (high activity level, 10 years registration period); Target 3 (medium activity level, 5 years registration period); Target 4 (low activity level, 1 month registration period); and Target 5 (low activity level, 2 years registration period). Comparing the attribute information of these 5 targets with the first screening criterion, it is found that: Target 1's account registration period does not meet the requirement; Target 4's activity level and account registration period both do not meet the requirement; Target 5's activity level does not meet the requirement; and Targets 2 and 3 both meet the requirement. Therefore, Targets 2 and 3 from the first screening are selected as the target targets.

[0336] Through steps 1710-1730 above, this embodiment combines keyword matching and audience screening to select target objects from the target set. Compared to related technologies that use a single indicator for matching or manual selection, this method can better match objects that meet the targeting requirements. Furthermore, combining keyword matching and audience screening to select target objects allows for personalized selection, ensuring that the selected target objects better meet the actual needs of the target audience, and improving the accuracy and rationality of the target object selection.

[0337] Since the first filtering criteria are often set by the first target based on their habits and preferences, this is often done without the first target being aware of the overall attribute distribution of the filtered objects. This can lead to a smaller number of target objects being selected, affecting the scope and breadth of content delivery and hindering the exposure of the first content. Therefore, this disclosure provides a solution that displays the filtering results based on the first filtering criteria to the first target, allowing them to easily adjust the filtering criteria and further filter out target objects. This flexible adjustment of the filtering criteria makes the first target's settings more reasonable, resulting in a more reasonable number of target objects and ultimately enabling the first content to better reach interested target objects, thus improving the accuracy of content delivery.

[0338] The following is combined Figure 18 and Figure 19A -B provides a detailed description of the process of performing a second screening on the objects after the first screening to obtain the target object.

[0339] Step 1810: Based on the matching of the first filtering conditions and the attribute information of the first filtered objects, perform a second filtering on the first filtered objects to obtain the second filtered objects;

[0340] Step 1820: Obtain the attribute information of the objects after the second filtering;

[0341] Step 1830: Display the distribution map of the objects after the second filtering based on attribute information;

[0342] Step 1840: Display the attribute information checkbox to receive the second filter condition input by the first object based on the distribution map;

[0343] Step 1850: Based on the matching of the second filtering conditions and the attribute information of the objects after the second filtering, a third filtering is performed on the objects after the second filtering to obtain the target object.

[0344] Steps 1810-1850 are described in detail below.

[0345] In step 1810, based on the matching of the first filtering condition with the attribute information of the first filtered object, a second filtering is performed on the first filtered object to obtain the second filtered object. The specific implementation process is similar to step 1730 above. The difference is that the second filtering in step 1810 yields the second filtered object, while the second filtering in step 1730 yields the target object. To save space, it will not be described in detail.

[0346] In step 1820, since the attribute information of each object is clearly recorded in the log data of the content delivery platform, the target terminal can obtain the attribute information of each object from the log data of each of the second-filtered objects in the content delivery platform. Specifically, the target terminal first sends an attribute information retrieval request to the content delivery platform; then, the content delivery platform retrieves the log data of each of the second-filtered objects from the backend server based on this request, and extracts the attribute information of each of the second-filtered objects from the log data. Finally, the content delivery platform sends the attribute information of each of the second-filtered objects back to the target terminal along with the response, thereby enabling the target terminal to obtain the attribute information of each of the second-filtered objects.

[0347] In step 1830, the total number of objects after the second screening is first counted. Next, the number of objects corresponding to each attribute item under each attribute information is counted, and the quotient of the number of objects to the total number of objects is calculated. Based on the quotient, the proportion of objects for each attribute item under each attribute information is determined. Finally, a distribution map of the objects after the second screening is generated based on the proportion of objects for each attribute item under each attribute information, and the distribution map is displayed to the first object on the front-end interface of the object terminal. The distribution map includes the distribution of the number of objects for each attribute information; this distribution map can include histograms, pie charts, etc., without limitation.

[0348] like Figure 19A As shown in the attribute distribution chart of the second-filtered objects, the total number of second-filtered objects is 160. The attribute distribution chart consists of three parts. The first part is a pie chart reflecting the activity level, where 100 second-filtered objects are highly active, and 60 are moderately active. The second part is a pie chart reflecting the account registration period, where 60 second-filtered objects have an account registration period of no more than 1 year, 80 have an account registration period between 1 and 5 years, and 20 have an account registration period of more than 5 years. The third part is a histogram reflecting the distribution of object tags, where 32 second-filtered objects contain tag 1, 48 contain tag 2, 16 contain tag 3, and 64 contain tag 4. When the attribute distribution of the second-filtered objects meets the expectations of the first-filter, click the "OK" button to complete the filtering of the target objects. When the first object needs to adjust the filtering conditions of the target object, the first object can click the "Check" button to input the second filtering conditions based on the specific situation of the attribute distribution map.

[0349] In step 1840, an attribute information checkbox is displayed on the front-end interface of the object terminal. This checkbox provides a selection option for the first object to input a second filter condition. When the first object selects the attribute information checkbox, the object terminal responds to this selection by displaying a settings interface for the second filter condition. The first object then inputs the specific content of the second filter condition on this interface. The second filter condition is used to match the attribute information of the objects after the second filter, and a third filter is then performed on these objects to obtain the target object.

[0350] like Figure 19BAs shown, after the first user clicks the "Check" button, the user's terminal displays the settings interface for the second filter condition. The interface shows input boxes for each filter item in the second filter condition, along with the prompt "Please enter the second filter condition." The filter items include user tags, user activity level, account registration years, and user region. The first user enters "Tag 1, Tag 4" in the user tag input box, "≥Medium Activity" in the user activity input box, ">1 year" in the account registration years input box, and "City A" in the user region input box. After completing the input for each filter item, the first user clicks the "OK" button to confirm the completion of the second filter condition settings.

[0351] In step 1850, based on the matching of the second filtering conditions and the attribute information of the objects after the second filtering, a third filtering is performed on the objects after the second filtering to obtain the target object. The specific implementation process is similar to step 1730 above. The difference is that step 1850 performs a third filtering on the objects after the second filtering based on the second filtering conditions to obtain the target object, while step 1730 performs a second filtering on the objects after the first filtering based on the first filtering conditions to obtain the target object. To save space, it will not be described in detail.

[0352] Through steps 1810-1850 above, this embodiment of the disclosure displays the distribution of attribute information of the second-filtered objects to the first object, enabling the first object to conveniently adjust the filtering conditions to obtain the second filtering conditions, and further filter out target objects based on the second filtering conditions. This method allows for flexible adjustment of filtering conditions, making the filtering conditions set by the first object more reasonable, thereby making the number of target objects more reasonable, and also better controlling the scope and breadth of content delivery, effectively increasing the exposure of the first content, enabling the first content to better reach interested target objects, and improving the accuracy of content delivery.

[0353] Because the matching degree of different objects after the second screening varies, the object tags they contain will differ significantly. When the number of objects after the second screening is large, displaying the object tags of each object after the second screening to the first object for verification and filtering is often time-consuming and inefficient in content delivery. This disclosure proposes a sampling display scheme based on matching degree, which can more reasonably display the object tags of the objects after the second screening to the first object, allowing the first object to select an appropriate number of retained objects. This improves the convenience of object display and the rationality of the selection of retained objects, and effectively improves the efficiency of content delivery.

[0354] The following is combined Figure 20 and Figure 21A-C provides a detailed description of the process of sampling and displaying based on matching degree.

[0355] Step 2010: Based on the matching of the filtering conditions with the attribute information of the first filtered object, perform a second filtering on the first filtered object to obtain the second filtered object;

[0356] Step 2020: Sort the objects after the second filtering according to their matching degree from high to low;

[0357] Step 2030: Select the second filtered object within the top two rankings as the first display object, and display the first display object and the object tag of the first display object that matches the target keyword;

[0358] Step 2040: Select a portion of the second-filtered objects from the first two positions in the ranking and use them as the second display objects. Display the second display objects and the object tags of the second display objects that match the target keywords.

[0359] Step 2050: Receive the number of reservations input by the first object;

[0360] Step 2060: Retain the second-filtered objects according to the retention number to obtain the target objects.

[0361] Steps 2010-2060 are described in detail below.

[0362] In step 2010, based on the matching of the filtering conditions with the attribute information of the first filtered objects, a second filtering is performed on the first filtered objects to obtain the second filtered objects. The specific implementation process is similar to step 1730 above. The difference is that the second filtering in step 2010 yields the second filtered objects, while the second filtering in step 1730 yields the target objects. To save space, it will not be described in detail.

[0363] In step 2020, based on the matching degree of each second-filtered object, the second-filtered objects are sorted in descending order according to their matching degree, resulting in a sorted list of the second-filtered objects. In this sorting, second-filtered objects with higher matching degrees are placed at the beginning, and second-filtered objects with lower matching degrees are placed at the end.

[0364] In step 2030, the specific values ​​for the top two rankings can be set according to actual circumstances and are not restricted. Specifically, firstly, the specific value for the second ranking is preset. Next, the second filtered object ranked in the top two is determined in the sorting process. Further, the second filtered object ranked in the top two is used as the first display object. Finally, the object tags in the object tags of the first display object that match the target keyword are determined, and the first display object and its object tags that match the target keyword are displayed to the first object on the front-end interface.

[0365] like Figure 21A As shown, the top three filtered objects in the ranking are selected as the first display objects. These first display objects include first display object 1, first display object 2, and first display object 3. The matching degree of first display object 1 is 0.96, that of first display object 2 is 0.87, and that of first display object 3 is 0.77. Next, the object tags in the object tags of first display object 1 that match the target keyword are identified as tags 1, 2, and 3; the object tags in the object tags of first display object 2 that match the target keyword are identified as tags 1 and 3; and the object tags in the object tags of first display object 3 that match the target keyword are identified as tags 1 and 3. Each first display object and its matching object tags are then displayed to the first display object.

[0366] In step 2040, the number of second-filtered objects selected from those never ranked in the top two positions is first determined. Next, based on this predetermined number, this number of second-filtered objects is randomly selected from those never ranked in the top two positions as the second display objects. Finally, the object tags of the second display objects that match the target keyword are determined, and the second display objects and their matching object tags are displayed to the first object on the front-end interface.

[0367] like Figure 21AAs shown, firstly, the number of second-filtered objects selected from those not ranked in the top two positions is determined to be 6. Next, six second-filtered objects are randomly selected from those not ranked in the top two positions as second-display objects. These second-display objects include second-display object 1 with a match score of 0.5, second-display object 2 with a match score of 0.42, second-display object 6 with a match score of 0.45, second-display object 4 with a match score of 0.62, second-display object 5 with a match score of 0.33, and second-display object 6 with a match score of 0.67. Next, the object tag matching the target keyword for second-display object 1 is determined as tag 1, the object tag matching the target keyword for second-display object 2 is determined as tag 2, the object tag matching the target keyword for second-display object 3 is determined as tag 1, the object tag matching the target keyword for second-display object 4 is determined as tag 3, the object tag matching the target keyword for second-display object 5 is determined as tag 1, and the object tag matching the target keyword for second-display object 6 is determined as tag 3. Finally, each second-display object and its matching object tag are displayed to the first object.

[0368] In step 2050, the number of objects to be retained is selected by the first object based on the first displayed object, the object tags of the first displayed object that match the target keyword, the second displayed object, and the object tags of the second displayed object that match the target keyword. Specifically, the first object inputs the number of objects to be retained on the front-end interface based on each of the displayed first displayed object, second displayed object, and each object tag. Then, the object terminal receives the number of objects to be retained input by the first object. This number of objects to be retained is the number of objects that the first content will target.

[0369] like Figure 21B As shown, the target terminal displays an input box for the number of people to retain and an "OK" button on the interface. The first target enters the number to retain, "100," on the interface and clicks the "OK" button. At this time, the target terminal receives the number of people to retain entered by the first target and sets the target target's number of people to 100.

[0370] In step 2060, the second-filtered objects are retained according to the number to be retained. The specific situations in which the target objects are obtained include, but are not limited to, the following two:

[0371] Scenario 1: Based on the number of items to be retained, select the second filtered object that ranks first in the sorting based on the number of items to be retained as the target object;

[0372] Scenario 2: Based on the number of objects to be retained, randomly select the number of objects to be retained after the second screening as the target objects.

[0373] like Figure 21CAs shown, the target terminal receives 100 retained objects. Based on a ranking determined by matching degree, the top 100 objects after the second filtering are selected as target objects. The final list of target objects is displayed on the front-end interface, including the target objects and the matching degree of each target object. The first object clicks the "OK" button, thus confirming the target objects for the first content delivery.

[0374] Through steps 2010-2060 above, this embodiment of the disclosure displays the second-ranked, most highly matched, second-filtered objects as the first display objects, one by one, to the first object. This clearly reflects the specific category of the object tags contained in the second-filtered objects with high matching scores. Furthermore, for the second-filtered objects whose matching scores are not ranked in the top two, a sampling method is used to randomly select a portion of the second-filtered objects as the second display objects. This significantly improves the convenience and rationality of object display, thereby increasing the efficiency of content delivery. It also allows the first object to determine a reasonable number of objects to retain by combining the first and second display objects, improving the rationality of the number of people targeted.

[0375] Another implementation of the content delivery method of this disclosure

[0376] After delivering the first content to the target audience, the interest expressed by each target audience will vary due to their different object tags. If the first weight of the target keywords in the first content is set to a fixed value by default, it is often detrimental to the accurate delivery of the first content. Therefore, this disclosure proposes a scheme that dynamically adjusts the first weight of the target keywords based on the varying degrees of interest each target audience has in the first content. This enables iterative optimization of the first weight of the target keywords, thereby improving the accuracy of content delivery.

[0377] The following is combined Figure 22 Another specific implementation process of the content delivery method according to the embodiments of this disclosure will be described in detail.

[0378] Step 310: Based on the number of published content of the same type as the first content to be delivered and the number of published content of the first object to which the first content is delivered, determine that the first condition is met;

[0379] Step 320: In response to the first condition being met, determine the first keyword based on the first content, expand the second keyword based on the first keyword, and integrate the first keyword and the second keyword into the target keyword;

[0380] Step 330: Obtain the object tags of each object in the object set on the content delivery platform;

[0381] Step 340: Select target objects from the object set based on the matching degree between target keywords and object tags;

[0382] Step 350: Deliver the first content based on the selected target audience;

[0383] Step 2210: Obtain the interest representation of the target audience after the first content is delivered;

[0384] Step 2220: Update the first weight of the target keywords in the first content based on the target audience's interest representation.

[0385] Steps 310-2220 are described in detail below.

[0386] Steps 310-350 have been described in detail above. Therefore, please refer to the above description for the specific process of steps 310-350 in the specific implementation of this embodiment. They will not be repeated here.

[0387] It should be noted that, in this embodiment of the disclosure, when the number of published content of the same type as the first content to be published and the number of published content of the first object to publish the first content are obtained, the individual permission or individual consent of the first object will be obtained through pop-up windows or redirection to a confirmation page, etc. After the individual permission or individual consent of the first object is clearly obtained, the number of published content of the same type as the first content to be published and the number of published content of the first object to publish the first content are obtained.

[0388] In addition, when obtaining the object tags of each object in the object set on the content delivery platform, the platform will also obtain the individual permission or consent of the first object through pop-up windows or redirection to a confirmation page. Only after obtaining the individual permission or consent of the first object will the platform obtain the object tags of each object in the object set on the content delivery platform.

[0389] In step 2210, after the first content is delivered to the target audience of the content delivery platform, the platform's backend server records the behavioral data of each target audience towards the first content. This behavioral data includes data generated by actions such as clicking, saving, liking, commenting, or forwarding. Next, the platform's backend server generates an interest representation for each target audience based on their behavioral data and sends this representation to the target audience's terminal. This interest representation indicates whether the target audience is interested in the first content. Specifically, when generating the interest representation based on the behavioral data, a judgment condition can be set first. Then, based on the matching of the judgment condition with the behavioral data, the target audience's interest representation is determined as either interested or uninterested.

[0390] For example, the judgment condition could be set as "if the target object has at least one of the following behaviors: saving, liking, or forwarding the first content, then the target object is determined to be interested in the first content." Then, based on the behavioral data of each target object, if at least one of the following behaviors is present in the behavioral data, the target object's interest is considered to be in the first content. If no such behaviors are present in the behavioral data, the target object's interest is considered to be indifferent to the first content.

[0391] In step 2220, the first weight is the ratio of the first number to the second number. The first number is the number of times the target object expressed interest after being presented with the first content containing the target keyword, and the second number is the number of times the first content containing the target keyword was presented to the target object. Specifically, when updating the first weight of the target keyword in the first content based on the target object's interest expression, target keywords in the first content that match the object tags of target objects who expressed interest in the first content are obtained, and the first weight of the obtained target keywords is adjusted to a higher value. Since the sum of the first weights of all target keywords in the first content is 1, the first weights of target keywords in the first content that do not match the object tags of target objects who expressed interest in the first content are adjusted to a lower value.

[0392] Through the above steps 310-2220, the embodiments of this disclosure can conveniently determine whether the weights of each target keyword of the first content are reasonable based on the target object's interest in the first content. According to the different object tags of the target objects with different interest representations, the first weights of each target keyword can be adjusted and updated. The adaptive iterative optimization of the first weights of the target keywords can be conveniently achieved, thereby improving the accuracy of subsequent content delivery.

[0393] Another implementation of the content delivery method of this disclosure

[0394] After the first content is delivered to the target objects, the interest expressed by each target object will vary due to the different object tags of each target object. If the second weight of the object tags is set to a fixed value by default, it is often difficult to accurately filter out target objects interested in the first content, resulting in the first content failing to effectively reach interested target objects. Therefore, this disclosure proposes a scheme that dynamically adjusts the second weight of object tags based on the different levels of interest of each target object in the first content. This enables iterative optimization of the second weight of object tags, thereby improving the accuracy of the filtered target objects and allowing the first content to better reach interested target objects.

[0395] The following is combined Figure 23 Another specific implementation process of the content delivery method according to the embodiments of this disclosure will be described in detail.

[0396] Step 310: Based on the number of published content of the same type as the first content to be delivered and the number of published content of the first object to which the first content is delivered, determine that the first condition is met;

[0397] Step 320: In response to the first condition being met, determine the first keyword based on the first content, expand the second keyword based on the first keyword, and integrate the first keyword and the second keyword into the target keyword;

[0398] Step 330: Obtain the object tags of each object in the object set on the content delivery platform;

[0399] Step 340: Select target objects from the object set based on the matching degree between target keywords and object tags;

[0400] Step 350: Deliver the first content based on the selected target audience;

[0401] Step 2310: Obtain the interest representation of the target audience after the first content is delivered;

[0402] Step 2320: Update the second weight of the object label of the target object according to the interest representation of the target object.

[0403] Steps 310-2320 are described in detail below.

[0404] Steps 310-350 have been described in detail above. Therefore, please refer to the above description for the specific process of steps 310-350 in the specific implementation of this embodiment. They will not be repeated here.

[0405] The specific implementation process of steps 2310-2320 is similar to that of steps 2210-2220 described above. The difference lies in that steps 2310-2320 update the second weight of the target object's object tag based on the target object's interest representation, while steps 2210-2220 update the first weight of the target keyword of the first content based on the target object's interest representation. The objects being updated are different. For the sake of brevity, further details will not be provided.

[0406] The second weight is the ratio of the third number to the fourth number. The third number is the number of times the target object expressed interest after the first content was delivered to the target object with the object tag, and the fourth number is the number of times the first content was delivered to the target object with the object tag.

[0407] Through the above steps 310-2320, the embodiments of this disclosure can conveniently determine whether the second weights of each object tag of the target object are reasonable based on the target object's interest in the first content. According to the different object tags of the target objects with different interest representations, the second weights of each object tag of the target object can be adjusted and updated. The adaptive iterative optimization of the second weights of the object tags can be conveniently realized, thereby improving the accuracy of the selected target objects and enabling the first content to better reach the target objects of interest.

[0408] Another implementation of the content delivery method of this disclosure

[0409] Because different target objects have different attribute information after the first content is delivered, their expression of interest in the first content will vary. Furthermore, attribute information can, to some extent, reflect the rationality of the filtering conditions set during content delivery. Based on this, this embodiment of the disclosure considers a scheme that statistically analyzes the percentage of target objects showing interest under various attribute information, and displays this statistical percentage to the first object as a reference for subsequent delivery tasks. This improves the accuracy of content delivery in subsequent delivery tasks after the first content is delivered.

[0410] The following is combined Figure 24 , Figure 25 Another specific implementation process of the content delivery method according to the embodiments of this disclosure will be described in detail.

[0411] Step 310: Based on the number of published content of the same type as the first content to be delivered and the number of published content of the first object to which the first content is delivered, determine that the first condition is met;

[0412] Step 320: In response to the first condition being met, determine the first keyword based on the first content, expand the second keyword based on the first keyword, and integrate the first keyword and the second keyword into the target keyword;

[0413] Step 330: Obtain the object tags of each object in the object set on the content delivery platform;

[0414] Step 340: Select target objects from the object set based on the matching degree between target keywords and object tags;

[0415] Step 350: Deliver the first content based on the selected target audience;

[0416] Step 2410: Obtain the interest representation of the target audience after the first content is delivered;

[0417] Step 2420: Obtain the attribute information of the target object;

[0418] Step 2430: For each attribute information, determine the ratio of the target object corresponding to the attribute information that represents the interest;

[0419] Step 2440: Display the ratios corresponding to each attribute information.

[0420] Steps 310-2440 are described in detail below.

[0421] Steps 310-350 have been described in detail above. Therefore, please refer to the above description for the specific process of steps 310-350 in the specific implementation of this embodiment. They will not be repeated here.

[0422] In step 2410, the specific implementation process of obtaining the target object's interest representation after the first content is delivered is similar to step 2210 above. To save space, it will not be described in detail again.

[0423] In step 2420, the specific implementation process for obtaining the attribute information of the target object is similar to the specific implementation process for obtaining the attribute information of the first filtered object in step 1730. The difference is that step 2420 obtains the attribute information of the target object, while step 1730 obtains the attribute information of the first filtered object. For the sake of brevity, this will not be elaborated further.

[0424] In step 2430, the total number of all target objects can first be counted. Next, for each attribute, the total number of people in the target objects corresponding to that attribute who indicate interest and the total number of people who indicate disinterest are calculated. Finally, the total number of people in the target objects corresponding to that attribute who indicate interest is divided by the total number of objects to obtain the percentage of people who indicate interest in the target objects corresponding to that attribute.

[0425] like Figure 25As shown, the target object's attribute information includes activity level, registration period, and object tags. The target object attribute distribution chart consists of three parts. The first part is the activity level ratio distribution chart, where the ratio of target objects with low activity level and no interest in the first content is 12.5%, the ratio of target objects with low activity level but interest in the first content is 25%, and the ratio of target objects with high activity level and interest in the first content is 62.5%. The second part is the registration period ratio distribution chart, where the ratio of target objects with a registration period of 5 years or more but no interest in the first content is 12.5%, the ratio of target objects with a registration period between 1 and 5 years but interest in the first content is 37.5%, the ratio of target objects with a registration period between 1 and 5 years but no interest in the first content is 12.5%, and the ratio of target objects with a registration period of no more than 1 year but interest in the first content is 37.5%. The third part is a percentage distribution chart of the target tags. Among the target objects with tag 1, 10 people are interested in the first content and 10 people are not interested in the first content; among the target objects with tag 2, 20 people are interested in the first content and 10 people are not interested in the first content; among the target objects with tag 3, 30 people are interested in the first content and no one is not interested in the first content; among the target objects with tag 4, 5 people are interested in the first content and 15 people are not interested in the first content.

[0426] In step 2440, after representing the percentages of interest in the target objects corresponding to the attribute information and generating an attribute distribution map, the percentage distribution of each attribute information is displayed to the first object on the front-end interface. The percentages of each attribute information are used to filter the third content for the first object after the first content is delivered, making the target object selection for the third content more accurate, thereby improving the accuracy of content delivery in subsequent delivery stages.

[0427] Through the above steps 310-2440, this embodiment of the disclosure can analyze the ratio of interest in the target object for each attribute information and display the ratio to the first object. This can help the first object to filter out more accurate target objects in the third content delivery filtering after the first content is delivered, thereby improving the efficiency of the third content reaching the target objects of interest and also improving the content delivery accuracy of each delivery task after the first content is delivered.

[0428] The specific interaction process between the object and the object terminal in the content delivery method of this disclosure embodiment. The following reference Figure 26 The following provides a detailed and exemplary description of the implementation details of the content delivery method according to the embodiments of this disclosure.

[0429] After receiving the first content to be delivered uploaded by the first object A, the content delivery recommendation system 141 in the object terminal 140 first extracts keywords based on the first content to obtain target keywords, as described in step 320 above. Next, it obtains the object tags of each object in the object set on the content delivery platform and matches the target keywords with the object tags. Based on the matching results, it filters out a preliminary target audience from the object set, as described in steps 330 and 1710 above. Then, it performs audience filtering on the preliminary target audience to obtain preliminary audience targeting and packet capture results, as described in step 1810 above. Further, in the analysis and filtering stage, it analyzes the preliminary audience targeting and packet capture results. This analysis includes audience size analysis, basic attribute analysis, scene activity analysis, object sampling analysis, etc. The analysis results are displayed to the first target audience (A), allowing A to input new filtering criteria or remove matched targets. Next, after A sets the new filtering criteria, a second audience filtering process is performed based on the new criteria to obtain the final target audience and packet capture results. The specific implementation process is described in steps 1820 to 1850 above. Finally, the target terminal 140 delivers the first content to the target audience of the content delivery platform via online delivery.

[0430] Furthermore, after completing the online campaign, the campaign volume, comparison of campaign metrics, and campaign performance are displayed to the first target audience. The campaign results are then integrated into the content recommendation system 141. Based on the level of interest of each target audience member in the campaign results towards the first content, the first weight of the target keywords and the second weight of the target tags are updated, thereby achieving iterative optimization of the first and second weights. The specific implementation process is described in steps 2220 and 2320 above. Simultaneously, based on the level of interest of each target audience member in the campaign results towards the first content, the interest ratio of each target audience member under each attribute information is generated and displayed to the first target audience. This ratio serves as the basis for setting audience selection criteria for the first target audience in subsequent content campaigns. The specific implementation process is described in steps 2430 to 440 above.

[0431] The following reference Figure 27 This document describes in detail a specific usage process of the content delivery method according to an embodiment of the present disclosure, including but not limited to steps 2701 to 2722.

[0432] Step 2701: The first object A sends a content delivery request to the object terminal 140;

[0433] Step 2702: The target terminal 140 obtains the number of published content of the same type as the first content in the content delivery request and the number of published content of the first object delivering the first content;

[0434] Step 2703: The target terminal 140 determines that the first condition is met based on the number of published content of the same type as the first content and the number of published content of the first object that delivered the first content.

[0435] Step 2704: The target terminal 140 determines the first keyword based on the text description of the first content, the text description of the second content delivered to the first object before the first content, and the multimodal parsing of the first content;

[0436] Step 2705: The target terminal 140, based on inputting the first keyword into the related word prediction model, expands the first keyword using a knowledge graph to obtain the second keyword;

[0437] Step 2706: The target terminal 140 integrates the first keyword and the second keyword into the target keyword, and obtains the target keyword and the first weight of the target keyword;

[0438] Step 2707: The object terminal 140 sends an object tag retrieval request to the content delivery platform server;

[0439] Step 2708: The content delivery platform server 110 extracts the object tags of each object from the object tag library 150 according to the object tag request, and sends the object tags of each object in the object set to the object terminal 140.

[0440] Step 2709: The target terminal 140 determines the matching degree between the target keyword and the object tag based on the target keyword, the first weight, the object tag, and the second weight.

[0441] Step 2710: Based on the matching degree between the target keywords and the object tags, the target terminal 140 performs a first screening in the object set to obtain the first-screened objects, and forms the first-screened objects into a preliminary screening population.

[0442] Step 2711: The target terminal 140 obtains the first filtering condition, filters the target audience based on the first filtering condition, obtains the second filtered target, and forms the initial target audience by combining the first filtered target.

[0443] Step 2712: The target terminal 140 obtains the attribute information of the initially selected target audience and generates an attribute information distribution map of the target audience.

[0444] Step 2713: The target terminal 140 displays the attribute information distribution map of the target audience to the first target.

[0445] Step 2714: The first object A sends the second filtering condition based on the distribution map selection to the object terminal 140;

[0446] Step 2715: The target terminal 140 performs a third screening on the initially selected target audience based on the second screening criteria to obtain the target audience and forms the final target audience.

[0447] Step 2716: The target terminal 140 displays the final target audience to the first target A;

[0448] Step 2717: The first object A sends a content delivery confirmation instruction to the object terminal 140;

[0449] Step 2718: The target terminal 140 delivers the first content to the target object on the content delivery platform server 110;

[0450] Step 2719: The content delivery platform server 110 sends the target object's interest expression to the target terminal 140;

[0451] Step 2720: Based on the interest representation of the target object, the target terminal 140 updates the second weight of the object tag of the target object and the first weight of the target keyword of the first content;

[0452] Step 2721: The object terminal 140 obtains the attribute information of the target object and determines the ratio of interest representation corresponding to each attribute information based on the attribute information;

[0453] Step 2722: Object terminal 140 displays the ratio corresponding to the attribute information to the first object A.

[0454] It is understood that the specific processes of steps 2701-2703 are similar to step 310 in the above embodiments. The specific processes of steps 2704-2706 are similar to step 320 in the above embodiments. The specific processes of steps 2707-2708 are similar to step 330 in the above embodiments. The specific processes of steps 2709-2716 are similar to step 340 in the above embodiments. The specific processes of steps 2717-2718 are similar to step 350 in the above embodiments. The specific process of step 2719 is similar to step 2210 in the above embodiments. The specific process of step 2720 is similar to steps 2220 and 2320 in the above embodiments. The specific processes of steps 2721-2722 are similar to steps 2420-2440 in the above embodiments. To save space, they will not be described in detail here.

[0455] Description of apparatus and devices according to embodiments of this disclosure

[0456] It is understood that although the steps in the above flowcharts are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated in this embodiment, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the above flowcharts may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages in other steps.

[0457] It should be noted that in various specific embodiments of this application, when processing is required based on data related to the characteristics of the target object, such as target object attribute information or a set of attribute information, the permission or consent of the target object will be obtained first. Furthermore, the collection, use, and processing of this data will comply with relevant laws, regulations, and standards. In addition, when embodiments of this application require obtaining target object attribute information, separate permission or consent from the target object will be obtained through pop-ups or redirection to a confirmation page. Only after obtaining the target object's separate permission or consent will the necessary target object-related data for the normal operation of the embodiments of this application be obtained.

[0458] Figure 28 A schematic diagram of the structure of a content delivery device 2800 provided in an embodiment of this disclosure. The content delivery device 2800 includes:

[0459] The condition determination unit 2810 is used to determine that the first condition is met based on the number of published content of the same type as the first content to be delivered and the number of published content of the first object to which the first content is delivered.

[0460] Keyword determination unit 2820 is used to determine a first keyword based on the first content in response to the first condition being met, expand a second keyword based on the first keyword, and integrate the first keyword and the second keyword into a target keyword.

[0461] Tag acquisition unit 2830 is used to acquire object tags of each object in the object set on the content delivery platform;

[0462] The object selection unit 2840 is used to select target objects from the object set based on the matching degree between target keywords and object tags;

[0463] Content delivery unit 2850 is used to deliver the first content based on the selected target audience.

[0464] Optionally, the condition determination unit 2810 is specifically used for:

[0465] If the number of published contents of the first object to which the first content is delivered is less than the first threshold, then the first condition is determined to be met.

[0466] Determine the first type of the first content;

[0467] Determine the number of published content of the first type on the content delivery platform;

[0468] If the number of published content of the first type is less than the second threshold, then the first condition is determined to be true.

[0469] Optionally, the keyword identification unit 2820 is specifically used for:

[0470] Extract the first set of keywords from the text description of the first content;

[0471] Extract the second sub-keyword from the text description of the second content placed before the first content by the first object;

[0472] Based on the multimodal analysis of the first content, the third sub-keyword is extracted;

[0473] The first, second, and third sub-keywords are combined into the first keyword.

[0474] Optionally, the keyword identification unit 2820 is specifically used for:

[0475] Input the first keyword into the related word prediction model to obtain the fourth sub-keyword;

[0476] The first keyword was expanded using a knowledge graph to obtain the fifth sub-keyword;

[0477] The fourth and fifth sub-keywords are combined into the second keyword.

[0478] Optionally, the target keyword has the first weight, the object tag has the second weight, and the object selection unit 2840 includes:

[0479] A matching degree determination unit (not shown) is used to determine the matching degree between the target keyword and the object tag based on the target keyword, the first weight, the object tag, and the second weight.

[0480] The target object selection unit (not shown) is used to select a target object from the object set based on the matching degree of each object in the object set.

[0481] Optionally, the second weight includes a first sub-weight from the first source and a second sub-weight from the second source, and the matching degree determination unit (not shown) is specifically used for:

[0482] Based on the target keyword, first weight, object tag, and first sub-weight, determine the first matching degree corresponding to the first source;

[0483] Based on the target keyword, the first weight, the object tag, and the second sub-weight, determine the second matching degree corresponding to the second source;

[0484] The matching degree is determined based on the first matching degree and the second matching degree.

[0485] Optionally, the first keyword's first weight is a fixed weight, and the content delivery device further includes a weight determination unit (not shown), specifically used for:

[0486] The first weight of the second keyword is determined, where the first weight of the fourth sub-keyword is the confidence probability of the fourth sub-keyword output by the related word prediction model; the first weight of the fifth sub-keyword is determined based on the distance between the fifth sub-keyword and the first keyword in the knowledge graph.

[0487] Optionally, the matching degree determination unit (not shown) is specifically used for:

[0488] For each object in the object set, identify a target object tag that matches the target keyword from the object tags of that object.

[0489] For each target object tag, determine the product of the second weight of the target object tag and the first weight of the matched target keyword;

[0490] The matching degree is obtained by summing the products of the labels of each target object.

[0491] Optionally, the target object selection unit (not shown) is specifically used for:

[0492] In the object set, identify candidate objects with a matching degree greater than the third threshold;

[0493] Sort the candidate objects in descending order of matching degree;

[0494] Select the candidate objects that rank first in the sort as the target objects.

[0495] Optionally, the tag acquisition unit 2830 is specifically used for:

[0496] For each object in the object set, obtain the set of content that the object represents that you are interested in;

[0497] Retrieve the third keyword of the content in the content collection;

[0498] For each third keyword, determine the number of content items in the content set that contain the third keyword;

[0499] If the number of contents exceeds the fourth threshold, the third keyword will be used as the object tag for the object.

[0500] Optionally, the object selection unit 2840 includes:

[0501] The first filtering unit (not shown) is used to perform a first filtering on the object set based on the matching degree between the target keywords and the object tags, and obtain the objects after the first filtering.

[0502] A filter condition acquisition unit (not shown) is used to acquire the first filter condition of the first object;

[0503] The second filtering unit (not shown) is used to perform a second filtering on the first filtered objects based on the matching of the attribute information of the first filtered objects with the first filtering conditions, so as to obtain the target object.

[0504] Optionally, the second filtering unit (not shown) is specifically used for:

[0505] Based on the matching of the first filtering condition and the attribute information of the first filtered object, a second filtering is performed on the first filtered object to obtain the second filtered object.

[0506] Retrieve the attribute information of the objects after the second filtering;

[0507] Display the distribution of objects based on attribute information after the second filtering;

[0508] Display attribute information checkboxes to receive second filter conditions input by the first object based on the distribution map;

[0509] Based on the matching of the second filtering criteria with the attribute information of the objects after the second filtering, a third filtering is performed on the objects after the second filtering to obtain the target object.

[0510] Optionally, the second filtering unit (not shown) is specifically used for:

[0511] Based on the matching of the filtering conditions with the attribute information of the first filtered objects, a second filtering is performed on the first filtered objects to obtain the second filtered objects.

[0512] Sort the objects after the second filter according to their matching degree from high to low;

[0513] The second filtered object within the top two positions in the ranking is used as the first display object, and the first display object and the object tags of the first display object that match the target keyword are displayed.

[0514] Select a portion of the second-filtered objects that have never ranked in the top two positions in the sorting as the second display objects, and display the second display objects and the object tags of the second display objects that match the target keywords;

[0515] The first object receives the number of items to be retained, wherein the number of items to be retained is selected by the first object based on the first display object, the object tags of the first display object that match the target keyword, the second display object, and the object tags of the second display object that match the target keyword;

[0516] The target objects are obtained by retaining the number of objects selected after the second filtering.

[0517] Optionally, the content delivery device further includes a first update unit (not shown), which is used for:

[0518] Obtain the target audience's interest representation after the first content is delivered;

[0519] Based on the target audience's interest, update the first weight of the target keyword in the first content. The first weight is the ratio of the first number to the second number. The first number is the number of times the target audience expressed interest after the first content containing the target keyword was delivered to the target audience. The second number is the number of times the first content containing the target keyword was delivered to the target audience.

[0520] Optionally, the content delivery device further includes a second update unit (not shown), which is used for:

[0521] Obtain the target audience's interest representation after the first content is delivered;

[0522] Based on the target object's interest representation, update the second weight of the target object's object tag. The second weight is the ratio of the third number to the fourth number. The third number is the number of times the target object expressed interest after the first content was delivered to the target object with the object tag. The fourth number is the number of times the first content was delivered to the target object with the object tag.

[0523] Optionally, the content delivery device further includes a display unit (not shown), which is used for:

[0524] Obtain the target audience's interest representation after the first content is delivered;

[0525] Obtain the attribute information of the target object;

[0526] For each attribute, determine the percentage of the target object that represents the attribute;

[0527] Display the ratios corresponding to each attribute information, which are used to filter the delivery of the third content to the first object after the first content.

[0528] Reference Figure 29 , Figure 29 To illustrate the structural block diagram of a terminal for implementing the content delivery method of this embodiment, the terminal includes: a radio frequency (RF) circuit 2910, a memory 2915, an input unit 2930, a display unit 2940, a sensor 2950, ​​an audio circuit 3030, a wireless fidelity (WiFi) module 2970, a processor 2980, and a power supply 2990, among other components. Those skilled in the art will understand that... Figure 29 The terminal structure shown does not constitute a limitation on mobile phones or computers and may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0529] The RF circuit 2910 can be used to receive and transmit signals during information transmission or calls. In particular, it receives downlink information from the base station and processes it with the processor 2980; in addition, it transmits uplink data to the base station.

[0530] The memory 2915 can be used to store software programs and modules, and the processor 2980 executes various functional applications and data processing of the target terminal by running the software programs and modules stored in the memory 2915.

[0531] The input unit 2930 can be used to receive input numeric or character information, and to generate key signal inputs related to the settings and function control of the target terminal. Specifically, the input unit 2930 may include a touch panel 2928 and other input devices 2929.

[0532] Display unit 2940 can be used to display input or provided information, as well as various menus of the target terminal. Display unit 2940 may include display panel 2941.

[0533] Audio circuit 3030, speaker 3028, and microphone 3029 provide an audio interface.

[0534] In this embodiment, the processor 2980 included in the terminal can execute the content delivery method of the previous embodiment.

[0535] The terminals disclosed in this embodiment include, but are not limited to, mobile phones, computers, intelligent voice interaction devices, smart home appliances, vehicle terminals, and aircraft. The embodiments of this invention can be applied to various scenarios, including but not limited to data security, blockchain, data storage, and information technology.

[0536] Figure 30This is a partial structural block diagram of a server for implementing the content delivery method of this disclosure. The server can vary significantly due to different configurations or performance characteristics, and may include one or more Central Processing Units (CPUs) 3022 (e.g., one or more processors) and a memory 3029, and one or more storage media 2130 (e.g., one or more mass storage devices) for storing application programs 3042 or data 3044. The memory 3029 and storage media 3030 may be temporary or persistent storage. The program stored in the storage media 3030 may include one or more modules (not shown in the diagram), each module including a series of instruction operations on the server. Furthermore, the CPU 3022 may be configured to communicate with the storage media 3030 and execute the series of instruction operations in the storage media 3030 on the server.

[0537] The server may also include one or more power supplies 3030, one or more wired or wireless network interfaces 3050, one or more input / output interfaces 3058, and / or one or more operating systems 3041, such as Windows Server™, Mac OS X™, Unix™, Linux™, FreeBSD™, etc.

[0538] The central processing unit 3022 in the server can be used to execute the content delivery method of the present disclosure embodiments.

[0539] This disclosure also provides a computer-readable storage medium for storing program code for executing the content delivery methods of the foregoing embodiments.

[0540] This disclosure also provides a computer program product comprising a computer program. A processor of a computer device reads and executes the computer program, causing the computer device to perform the aforementioned transaction on-chaining.

[0541] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in this disclosure and the foregoing drawings are used to distinguish similar objects and are not necessarily used to describe a particular 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, for example, in orders other than those illustrated or described herein. Furthermore, the terms “comprising” and “including,” and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatuses.

[0542] It should be understood that in this disclosure, "at least one item" means one or more, and "more than one" means two or more. "And / or" is used to describe the relationship between related objects, indicating that three relationships can exist. For example, "A and / or B" can represent three cases: only A exists, only B exists, and both A and B exist simultaneously, where A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one of the following" or similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one of a, b, or c can represent: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, and c can be single or multiple.

[0543] It should be understood that in the description of the embodiments disclosed herein, "multiple" means two or more, "greater than", "less than", "exceeding" etc. are understood to exclude the number itself, and "above", "below", "within" etc. are understood to include the number itself.

[0544] In the several embodiments provided in this disclosure, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces, indirect coupling or communication connection between apparatuses or units, and may be electrical, mechanical, or other forms.

[0545] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0546] Furthermore, the functional units in the various embodiments of this disclosure can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0547] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this disclosure, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this disclosure. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0548] It should also be understood that the various implementation methods provided in this disclosure can be combined arbitrarily to achieve different technical effects.

[0549] The above is a detailed description of the embodiments of this disclosure. However, this disclosure is not limited to the above embodiments. Those skilled in the art can make various equivalent modifications or substitutions without departing from the spirit of this disclosure. All such equivalent modifications or substitutions are included within the scope defined by the claims of this disclosure.

Claims

1. A content delivery method, characterized in that, include: Based on the number of published content of the same type as the first content to be delivered and the number of published content of the first object delivering the first content, the first condition is determined to be met; In response to the first condition being met, a first keyword is determined based on the first content, a second keyword is expanded based on the first keyword, and the first keyword and the second keyword are integrated as the target keyword; Obtain the object tags of each object in the object set on the content delivery platform; Based on the matching degree between the target keywords and the object tags, target objects are selected from the object set; The first content is delivered based on the selected target object; The determination that the first condition is met based on the number of published content of the same type as the first content to be delivered and the number of published content of the first object delivering the first content includes: If the number of published content of the first object that delivers the first content is less than a first threshold, then the first condition is determined to be true; if the number of published content of the first object that delivers the first content is greater than or equal to the first threshold, then the first type of the first content is determined, and the number of published content of the first type on the content delivery platform is determined; if the number of published content of the first type is less than a second threshold, then the first condition is determined to be true. The process of expanding the second keyword based on the first keyword includes: The first keyword is input into the related word prediction model, and the related word prediction model is used to perform semantic analysis and word prediction on the first keyword to obtain the related words corresponding to the first keyword. The related words are used as the fourth sub-keyword. The first keyword is expanded using a knowledge graph to obtain the fifth sub-keyword. The fourth sub-keyword and the fifth sub-keyword are integrated into the second keyword.

2. The content delivery method according to claim 1, characterized in that, The step of determining the first keyword based on the first content includes: Extract the first set of keywords from the text description of the first content; Extract the second sub-keyword from the text description of the second content placed before the first content by the first object; Based on the multimodal analysis of the first content, the third sub-keyword is extracted; The first sub-keyword, the second sub-keyword, and the third sub-keyword are integrated into the first keyword.

3. The content delivery method according to claim 1, characterized in that, The target keyword has a first weight, and the object tag has a second weight; The step of selecting target objects from the object set based on the matching degree between the target keywords and the object tags includes: Based on the target keyword, the first weight, the object tag, and the second weight, the matching degree between the target keyword and the object tag is determined; Based on the matching degree of each object in the object set, a target object is selected from the object set.

4. The content delivery method according to claim 3, characterized in that, The second weight includes a first sub-weight from a first source and a second sub-weight from a second source, wherein the first source and the second source are different source channels of the object tag; The step of determining the matching degree between the target keyword and the object tag based on the target keyword, the first weight, the object tag, and the second weight includes: Based on the target keyword, the first weight, the object tag, and the first sub-weight, a first matching degree corresponding to the first source is determined; Based on the target keyword, the first weight, the object tag, and the second sub-weight, a second matching degree corresponding to the second source is determined; The matching degree is determined based on the first matching degree and the second matching degree.

5. The content delivery method according to claim 3, characterized in that, The first weight of the first keyword is a fixed weight; After expanding the second keyword based on the first keyword, the content delivery method further includes: determining the first weight of the second keyword, wherein the first weight of the fourth sub-keyword is the confidence probability of the fourth sub-keyword output by the related word prediction model; the first weight of the fifth sub-keyword is determined based on the distance between the fifth sub-keyword and the first keyword in the knowledge graph.

6. The content delivery method according to claim 3, characterized in that, The step of determining the matching degree between the target keyword and the object tag based on the target keyword, the first weight, the object tag, and the second weight includes: For each object in the object set, determine a target object tag that matches the target keyword from the object tags of that object; For each target object tag, determine the product of the second weight of the target object tag and the first weight of the matching target keyword; The matching degree is obtained by summing the products of each target object label.

7. The content delivery method according to claim 3, characterized in that, The step of selecting a target object from the object set based on the matching degree of each object in the object set includes: In the object set, candidate objects with a matching degree greater than a third threshold are identified; The candidate objects are sorted in descending order of matching degree; The top five candidate objects in the sorted list are selected as the target objects.

8. The content delivery method according to claim 1, characterized in that, The object tags of each object in the object set on the content delivery platform include: For each object in the object set, obtain the set of content of interest represented by the object; Obtain the third keyword of the content in the aforementioned content set; For each of the third keywords, determine the number of contents in the content set that contain the third keyword; If the number of contents is greater than the fourth threshold, the third keyword will be used as the object tag of the object.

9. The content delivery method according to claim 1, characterized in that, The step of selecting target objects from the object set based on the matching degree between the target keywords and the object tags includes: Based on the matching degree between the target keywords and the object tags, a first screening is performed on the object set to obtain the first-screened objects; Obtain the first filtering condition for the first object; Based on the matching of the first filtering conditions with the attribute information of the first filtered objects, a second filtering is performed on the first filtered objects to obtain the target object.

10. The content delivery method according to claim 9, characterized in that, The step of matching the attribute information of the first filtered object with the first filtered criteria to perform a second filtering on the first filtered object to obtain the target object includes: Based on the matching of the first filtering condition with the attribute information of the first filtered object, a second filtering is performed on the first filtered object to obtain a second filtered object. Obtain the attribute information of the second filtered object; Display a distribution map of the objects after the second filtering based on the attribute information; Display an attribute information checkbox to receive a second filtering condition input by the first object based on the distribution map; Based on the matching of the second filtering conditions with the attribute information of the second filtered objects, a third filtering is performed on the second filtered objects to obtain the target object.

11. The content delivery method according to claim 9, characterized in that, The step of matching the attribute information of the first filtered object with the first filtered criteria to perform a second filtering on the first filtered object to obtain the target object includes: Based on the matching of the filtering conditions with the attribute information of the first filtered object, a second filtering is performed on the first filtered object to obtain a second filtered object. The objects after the second filtering are sorted from high to low according to the matching degree; The second filtered object in the top second position of the ranking is used as the first display object, and the first display object and the object tag of the first display object that matches the target keyword are displayed. Select a portion of the second filtered objects from those that have never ranked in the first second place in the sorting, and display them as the second display objects, along with the object tags of the second display objects that match the target keyword. The first object receives the number of objects to be retained, wherein the number of objects to be retained is selected by the first object based on the first display object, the object tags of the first display object that match the target keyword, the second display object, and the object tags of the second display object that match the target keyword. The target object is obtained by retaining the second filtered object according to the stated retention number.

12. The content delivery method according to claim 1, characterized in that, After delivering the first content based on the selected target object, the content delivery method further includes: Obtain the interest representation of the target audience after the first content is delivered; Based on the target object's interest representation, update the first weight of the target keyword in the first content, wherein the first weight is the ratio of a first number to a second number, the first number is the number of times the target object expressed interest after the first content containing the target keyword was delivered to the target object for the target keyword, and the second number is the number of times the first content containing the target keyword was delivered to the target object for the target keyword.

13. The content delivery method according to claim 1, characterized in that, After delivering the first content based on the selected target object, the content delivery method further includes: Obtain the interest representation of the target audience after the first content is delivered; Based on the interest representation of the target object, update the second weight of the object tag of the target object, wherein the second weight is the ratio of the third number to the fourth number, the third number is the number of times the target object expressed interest after the first content was delivered to the target object with the object tag, and the fourth number is the number of times the first content was delivered to the target object with the object tag.

14. The content delivery method according to claim 1, characterized in that, After delivering the first content based on the selected target object, the content delivery method further includes: Obtain the interest representation of the target audience after the first content is delivered; Obtain the attribute information of the target object; For each of the attribute information, determine the ratio of the target object corresponding to the attribute information that represents the interest; The ratios corresponding to each of the aforementioned attribute information are displayed for filtering the delivery of the third content of the first object after the first content is delivered.

15. A content delivery device, characterized in that, include: The condition determination unit is used to determine that the first condition is met based on the number of published content of the same type as the first content to be delivered and the number of published content of the first object delivering the first content. A keyword determination unit is used to, in response to the first condition being met, determine a first keyword based on the first content, expand a second keyword based on the first keyword, and integrate the first keyword and the second keyword into a target keyword. The tag acquisition unit is used to acquire the object tags of each object in the object set on the content delivery platform. An object selection unit is used to select a target object from the object set based on the matching degree between the target keyword and the object tag; A content delivery unit is used to deliver the first content based on the selected target object; The determination that the first condition is met based on the number of published content of the same type as the first content to be delivered and the number of published content of the first object delivering the first content includes: If the number of published content of the first object that delivers the first content is less than a first threshold, then the first condition is determined to be true; if the number of published content of the first object that delivers the first content is greater than or equal to the first threshold, then the first type of the first content is determined, and the number of published content of the first type on the content delivery platform is determined; if the number of published content of the first type is less than a second threshold, then the first condition is determined to be true. The process of expanding the second keyword based on the first keyword includes: The first keyword is input into the related word prediction model, and the related word prediction model is used to perform semantic analysis and word prediction on the first keyword to obtain the related words corresponding to the first keyword. The related words are used as the fourth sub-keyword. The first keyword is expanded using a knowledge graph to obtain the fifth sub-keyword. The fourth sub-keyword and the fifth sub-keyword are integrated into the second keyword.

16. An electronic device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the content delivery method according to any one of claims 1 to 14.

17. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the content delivery method according to any one of claims 1 to 14.

18. A computer program product comprising a computer program that is read and executed by a processor of a computer device, causing the computer device to perform the content delivery method according to any one of claims 1 to 14.

Images