TWI795707B - Content recommendation system and content recommendation method - Google Patents

Abstract

A content recommendation system and a content recommendation method are provided in the present disclosure. The content recommendation system includes a server that receives a plurality of user intention information from a terminal device. The terminal device includes a display module for displaying a web page group. The web page group includes a plurality of web page information. Each of the web page information is connected to another web page information. The server provides at least one user-recommended content based on multiple user intention information provided by the terminal device.

Description

Content Recommendation System and Content Recommendation Method

The present invention relates to a content recommendation system and a content recommendation method, in particular to a content recommendation system and a content recommendation method based on user intentions.

Most webpage information or APPs for mobile devices use the number of clicks on the content information as the basis for recommending content. However, if the user cannot find the desired product or service within a certain period of time, he will find another way, thus losing potential customers.

In addition, the existing technology often divides users into groups when recommending content, and the RFM model is often used for grouping, that is, users are grouped by Recency (date of last consumption), Frequency (consumption frequency), and Monetary (consumption amount) , however, this model cannot be applied to new customers due to the lack of consumption records, and the existing technology cannot judge user intentions, which greatly reduces the effect of recommendation.

Therefore, how to provide a content recommendation system and a content recommendation method that can recommend content information according to the user's intent has become one of the important issues to be solved by this project.

The technical problem to be solved by the present invention is to provide a content recommendation system for the deficiencies of the prior art, including: a server receiving a plurality of user intention information of a terminal device, the terminal device includes a display module for displaying web pages A group, the webpage group includes a plurality of webpage information, each webpage information is at least connected to another webpage information; wherein, the server provides at least one user recommended content according to the plurality of user intention information provided by the terminal device.

In order to solve the above technical problems, one of the technical solutions adopted by the present invention is to provide a content recommendation method, including: providing a webpage group, wherein the webpage group includes a plurality of webpage information, and each webpage information is at least related to another Web page information connection; receiving a plurality of user intention information; calculating a user intention grouping information according to the plurality of user intention information and a deep learning model; and providing a recommended content according to the user intention grouping information.

One of the beneficial effects of the present invention is that the content recommendation system and content recommendation method provided by the present invention can effectively use the user's web browsing track information and attention information in the web page group to accurately provide users with appropriate recommended content , even for new customers with no consumption records, it can also increase users' high interest in website groups and related user experience.

In order to further understand the features and technical content of the present invention, please refer to the following detailed description and drawings related to the present invention. However, the provided drawings are only for reference and description, and are not intended to limit the present invention.

The following are specific examples to illustrate the implementation of the "content recommendation system and content recommendation method" disclosed in the present invention. Those skilled in the art can understand the advantages and effects of the present invention from the content disclosed in this specification. The present invention can be implemented or applied through other different specific embodiments, and various modifications and changes can be made to the details in this specification based on different viewpoints and applications without departing from the concept of the present invention. In addition, the drawings of the present invention are only for simple illustration, and are not drawn according to the actual size, which is stated in advance. The following embodiments will further describe the relevant technical content of the present invention in detail, but the disclosed content is not intended to limit the protection scope of the present invention. In addition, the term "or" used herein may include any one or a combination of more of the associated listed items depending on the actual situation.

[first embodiment]

Please refer to FIG. 1 and FIG. 2 . FIG. 1 is a schematic diagram of a content recommendation system according to a first embodiment of the present invention. FIG. 2 is a schematic diagram of a web page group in the present invention.

In this embodiment, the content recommendation system SYS1 includes a server 1 . The server 1 can receive a plurality of user intention information of the terminal device CD. The terminal device CD includes at least one display module (not shown in the figure) for displaying a web page group WPG. The webpage group WPG includes a plurality of webpage information P1-P10. Each web page information P1-P10 is at least connected to another web page information. In addition, each webpage information P1-P10 includes a plurality of area information. That is, each web page information P1-P10 can be divided into different areas, for example: the web page information P1 includes the first area information P11, the second area information P12 and the third area information P13. The first region information, the second region information P12 and the third region information P13 are arranged consecutively. The display module (not shown) of the terminal device CD can determine which area of the webpage information P1-P10 it is according to the currently displayed area.

Please refer to FIG. 2 , the webpage group WPG includes ten webpage information P1-P10. The webpage information P1 is the initial webpage of the website, and the other webpage information P2, webpage information P5, and webpage information P8 are links to the webpage information P1. That is, each web page information will link at least one other web page information. In this embodiment, in addition to the basic information, initial recommended content can be set for the user to choose.

In this embodiment, the server 1 can provide at least one user recommended content according to a plurality of user intention information provided by the terminal device CD. User recommended content is recommended content provided based on user intent information. The user recommended content may be the same as or different from the initial recommended content.

The user intention information at least includes webpage browsing track information and attention degree information. Wherein, the webpage browsing trace information includes a browsing trace of a plurality of webpage information P1-P10. The attention degree information at least includes a dwell time of each web page information P1-P10.

The webpage browsing track information is the track of the user viewing the plurality of webpage information P1-P10 in the webpage group WPG. For example, the user starts to browse from the webpage information P1 first, then clicks to enter the webpage information P2, then enters the webpage information P3, and then enters the webpage information P4. At this time, the user's webpage browsing track information is: "webpage information P1, webpage information P2, webpage information P3, webpage information P4". Other webpage browsing track information WP2 can be "webpage information P1, webpage information P8, webpage information P9, webpage information P10".

Furthermore, the webpage browsing track information can also consider the different regions where the user viewed the webpage information P1-P10, for example, the multiple regions information P11-P13 of the webpage information P1. For example, start browsing from the first area information P11, then the third area information P13, and then the second area information P12. In other embodiments, a gaze tracking technology of the terminal device CD may be used to track the user's gaze, so as to more accurately obtain the user's browsing track in the webpage group WPG.

The attention degree information at least includes the user's stay time and number of visits on each webpage information P1-P10. In addition, the attention degree information may also include user's click information on each webpage information P1-P10. For example, click on the webpage information P3.

Please refer to FIG. 3 , FIG. 4 and FIG. 5 . FIG. 3 is a schematic diagram of the content recommendation system of the present invention calculating attention information using a deep learning model. FIG. 4 is a schematic diagram of calculating user intention grouping information by the content recommendation system of the present invention using a deep learning model. FIG. 5 is another schematic diagram of the content recommendation system of the present invention using a deep learning model to calculate user intention grouping information.

Then, the server 1 calculates user intention grouping information through a deep learning model according to the user's web browsing track information and attention information on each web page information P1-P10 of the web page group WPG. The server 1 then provides user recommended content according to user intention grouping information and a plurality of content information. Wherein, the user intention grouping information may include low-level intention information, middle-level intention information and high-level intention information. The low-level intention information is "only browsing the website, without obvious purchase intention"; the middle-level intention information is "the intention to purchase products or services, but the purchase items are uncertain"; is the intent of the scheduled service". That is to say, the present invention judges the user's intention to purchase a product or service based on the user's various behaviors of browsing webpage information and user information.

In this embodiment, a necessary information and a non-essential information are used to perform the calculation of the deep learning model, so as to obtain the aforementioned attention degree information. Wherein, the deep learning model is, for example but not limited to, a cyclic network model (Recurrent neural network, RNN), a long short-term memory model (Long Short-Term Memory, LSTM) in a time cyclic network model, or a time cyclic network model The bidirectional long short-term memory model (Bidirectional Long Short-Term Memory, BiLSTM) in the road model.

As shown in FIG. 3 , the necessary information includes: the page types of web page information P1-P10, the length or ratio of the user's stay when browsing each web page information P1-P10, the number or ratio of views, and the viewing date. Non-essential information includes: gender, company, country and other information in user information. The recommender system of the present invention calculates the relationship between different webpage information by means of a cyclic network model, for example: webpage information P1 and webpage information P2, which are referred to as attention information here, for example, users often use webpage information P1 and webpage information Watching back and forth between P1 and its attention will be close to 1. The server 1 of the content recommendation system SYS1 determines that the webpage information P1 has a strong relationship with the webpage information P1. Conversely, if the user reads the webpage information P1 but has never seen the webpage information P2, then the attention information between the two will be 0, and it will be determined that the webpage information P1 and the webpage information P1 have no relationship. The road model is used to calculate the attention information of each web page information P1-P10.

In addition, in this embodiment, the page types of the web page information P1-P10 can be preset with different weights for the calculation of the deep learning model. That is, each webpage information P1-P10 corresponds to a predetermined weight. The predetermined weights corresponding to each webpage information P1-P10 can be the same or different. In this embodiment, the predetermined weight corresponding to the webpage information P1 which is the initial webpage information of the webpage group WPG is the lowest. Because most of the initial web pages will not include product information or service information. In other embodiments, the web page information P1 as the initial web page can be provided with recommended product content or recommended service content for users to browse, so as to increase the user's purchase intention. The predetermined weight of each webpage information P1-P10 can be set according to the setting position of the webpage information P1-P10, and the predetermined weight can also be set according to the product content or service content included in each webpage information P1-P10. In this embodiment, the webpage information P1-P10 is set at the initial location of the webpage group WPG, that is, the entry webpage information, and the predetermined weight of the webpage information P1 can be set to 0.1. However, if the recommended content in the webpage information P1 is clicked on, the predetermined weight of the webpage information P1 can be slightly adjusted, for example adjusted to 0.2. When the webpage information setting position is the middle position of the webpage group WPG, for example: webpage information P2-P3, webpage information P5-P6, etc., the predetermined weight of webpage information P2-P3 and webpage information P5-P6 can be set to 0.3- 0.4. When the product content or service content of a predetermined webpage information has high value, for example, the webpage information P5 includes a newly launched product or service, the reservation weight of the webpage information P5 can be increased.

Specifically, if user A's user intention information is: enter the portal webpage and stay for a period of time, then click to enter the consumer electronics product webpage in the webpage group WPG, and then click to enter the lower-level home appliances and notebook computers respectively. products such as electronics and mobile phones, and frequently switch back and forth between consumer electronics product pages and lower-level product pages. Shopping cart, this kind of behavior of "only browsing the webpage, unable to clearly distinguish the purchase intention of the product" can still be detected by the present invention and classified as "low-level intention".

User B’s user intention information is: After clicking from the entrance webpage to enter the consumer electronics product webpage, and then continuously clicking back and forth between different brands of mobile phone product webpages, although there is no significant difference in the proportion of the dwell time of each webpage, there is no difference. Put related products into the shopping cart, but user B’s interest range has been narrowed down to mobile phone products, but the brand or model cannot be confirmed. This type of product has the intention of purchasing products or services, but does not have an exact target The behavior of "product or service" can still be detected by the present invention and classified as "intermediate intent".

The user intention information of user C is: enter the consumer electronics product webpage from the entrance webpage and then click on the mobile phone product webpage, and then spend more time on the webpage of a certain brand and model of mobile phone products than other webpages , although the relevant products were not put into the shopping cart in the end, but to further narrow the scope of interest of user C to a certain model of mobile phone of a certain brand, this kind of behavior of "having a clear intention to purchase products or services" can be recognized The present invention detects and classifies as "higher order intent".

Therefore, although user A, user B, and user C have not put the product into the shopping cart, the content recommendation system SYS1 of this embodiment can group them to facilitate subsequent product recommendation. Among them, attention information can be calculated by the following formula:

DL(NIN, UNIN , PW, PN) = ATT (Equation 1)

Among them, NIN is necessary information. UNIN is non-essential information. PW is a predetermined weight of web page information. PN is the number of page views, and the DL function is calculated according to a deep learning model.

As shown in Figure 4, the deep learning model is calculated using web browsing track information and attention information to calculate user intention grouping information. The fully connected layer deep learning model includes an input layer, multiple hidden layers, and an output layer. Each web page browsed contains at least one product type. Assuming that the user browses the home-use network storage device in the special offer page, the number of visits for the special offer and the home-use product will be increased by one each, and the input layer of the model is the web page browsing track information And the degree of attention, such as but not limited to the time and frequency of browsing home product webpages, the time and frequency of enterprise product webpage browsing, the time and frequency of accessory product webpage browsing, and the degree of attention between the product and the current page. The output layer is the probability vector of each group of user intentions, and the highest probability is taken as the intention of the application. For example, the probability vector of user intentions is (C1, C2 ,C3)= (0.3, 0.7, 0.8), then choose C3 : Among them, C1 refers to only browsing the website without obvious purchase intention; C2 refers to the intention to purchase products or services, but the purchase items are uncertain; C3 refers to the definite intention to purchase predetermined products or predetermined services. Fig. 5 is to convert the deep learning model in Fig. 4 into a flowchart for presentation.

According to Figure 5, the following formula can be obtained:

ML(WT, ATT) = UIN1 (Equation 2)

Wherein, UIN1 is user intent information. WT is web browsing track information, including the order in which users browse web pages. ATT is attention information. Attention information includes: the dwell time of each web page information, the number of views, and the predetermined weight. The ML function uses a supervised machine learning to calculate user intention information.

Please refer to FIG. 6 and FIG. 7, FIG. 6 is a functional block diagram of the content recommendation system of the present invention. FIG. 7 is a schematic diagram of the content recommendation system of the present invention providing user recommended content according to user intention grouping information.

The terminal device CD will firstly display the web page group WPG. As mentioned above, the webpage group WPG will first display an initial recommended content for user reference, wherein the initial recommended content is for example but not limited to a hot product or service, a promotional product or service, or a newest product or service. The terminal device CD will send the user information, the webpage browsing track information and attention degree information of the user browsing webpage group WPG to the server 1 .

In this embodiment, the terminal device CD includes a terminal communication module (not shown). The terminal communication module (not shown in the figure) can be a wired communication module or a wireless communication module.

The server 1 includes an analysis module 11 , a storage module 12 and a communication module 13 . The analysis module 11 is electrically connected to the storage module 12 and the communication module 13 .

The terminal device CD can be a desktop computer, a notebook computer, a smart phone, a tablet computer or a wearable electronic device.

The analysis module 11 is a central processing unit (CPU), an application specific integrated circuit (ASIC) or a microprocessor (MCU). The storage module 12 is a flash memory, a read-only memory, a programmable read-only memory, an electrically rewritable read-only memory, an erasable programmable read-only memory or an electronically programmable Erase configurable read-only memory.

The communication module 13 may include a wired communication unit (not shown) and a wireless communication unit (not shown). The wired communication unit (not shown) can also be set independently to connect with other databases. When the communication module 13 is a wireless communication unit, the communication module 13 can be a Wi-Fi communication unit, a Bluetooth communication unit, a Zigbee communication unit (Zigbee), a LoRa communication unit, a Sigfox communication unit or A NB-IoT communication unit.

The server 1 receives the user information through the communication module 13, the webpage browsing track information of the user browsing webpage group WPG, and the attention degree information are sent to the server 1, and the server 1 will further use the analysis module 11 to analyze the user information. . The user browses webpage browsing track information and attention information of the webpage group WPG, and uses a deep learning model to calculate the attention information.

At this time, the terminal device CD will firstly send the user login information to the server 1 . That is, the terminal device CD will send the information of whether the user has logged into the webpage group WPG to the server 1, for example: logged in (new member), logged in (old member), not logged in.

Furthermore, the analysis module 11 of the server 1 uses the web page browsing track information and attention degree information to calculate user intention grouping information using a type of neural model. The analysis module 11 of the server 1 will further provide a user recommended content to the terminal device CD for display according to the user intention grouping information and multiple content information. The content information may include product information or service information.

In this embodiment, the user intention information can be simplified into the following formula for calculation:

ML(WT , ATT , URD)= UIN2 (Equation 3)

Wherein, UIN2 is user intent information. WT is web browsing track information, including the order in which users browse web pages. ATT is attention information. The attention information includes: the dwell time of each web page information, the number of views, and the predetermined weight. URD is the user login information. The ML function is one of supervised machine learning to calculate user intent information.

When the user sees the displayed user recommended content, the web page group WPG of the terminal device CD can send a user preference information to the server according to whether the user clicks on the user recommended content or clicks on the dislike user recommended content 1. The user preference information is information about the user's preference or clicks on the user-recommended content. The analysis module 11 of the server 1 will further calculate the user login information, user preference information, web browsing track information and attention degree information through deep learning models such as deep learning, and calculate the user intention grouping information again. Then, the server 1 will provide another recommended content for the user according to the recalculated user intention grouping information and multiple content information. The analysis module 11 of the server 1 will use the communication module 13 to transmit the re-provided user recommendation content to the terminal device CD for display, and provide the calculated user recommendation content to the user again.

The analysis module 11 of the server 1 can calculate the content recommended by the user every predetermined time interval. In each predetermined time interval, the user's web browsing track information and attention degree information will change, so the user's recommended content will also be adjusted accordingly. The predetermined time interval can be adjusted according to actual needs, which is not limited in the present invention.

In this embodiment, the user recommended content may provide corresponding recommended content according to different user intention information. For example: when the user intent information is high-level intent information, the analysis module can provide different product content and service content matching the product content or service content selected by the user according to the product content or service content selected by the user, so as to further provide High value product or service.

When the user intention information is high-level intention information, and the user selects a network storage server (Network Attached Storage, NAS), the content recommendation system 1 can provide a high-speed wireless network card (Wi-Fi 6) related products , or cloud storage service content to enhance the service content of the network storage server.

When the user intention information is middle-level intention information, that is, the user has the intention to purchase a product or service, but the purchase item is not clear. At this time, the content recommendation system 1 can provide products or services similar to the product content or service content included in the webpage browsing track information to the user. Taking a network storage server as an example, web page browsing track information includes a network storage server with a medium-level specification. At this time, the content recommendation system 1 can provide users with a network storage server with a high-level specification and a medium-level specification. The user of the server recommends content to the user. That is, when the user intention information is a middle-level intention information, the user cannot decide which product content or service content to purchase from the browsed web page information, and the content recommendation system 1 will provide the user with a wider range of information. Product content or service content for selection.

When the user intention information is a low-level intention information, that is, the content recommendation system 1 can provide multiple product contents or service contents in the web page group WPG according to the specifications, prices, product types or service types. recommend content to customers. In addition, the content recommendation system 1 can also randomly recommend multiple product contents or service contents in the web page group WPG.

That is, the content recommendation system 1 provides different user recommended content according to different types of user intention information.

In addition, the server 1 stores the user login information, user preference information, web page browsing track information, attention degree information, etc. corresponding to the user in the storage module 12 . That is, to establish a database file for users as the basis for improving user experience and providing better services.

Please refer to FIG. 8 . FIG. 8 is another schematic diagram of the content recommendation system of the present invention providing user recommended content according to user intention grouping information.

The content recommendation system in FIG. 8 is similar to the content recommendation system in FIG. 7 , and the server 1 can also collect the first user information of the user to further enhance the analysis accuracy of the user intention grouping information. The first user information is various identity information provided by the user in the webpage group WPG, such as photos, personal information, and the like. In addition, the first user information can also include information on other cooperative websites or cooperative organizations, as well as de-identified information in other databases, such as age, interests, and place of residence. Identifying user information (user A, 40-year-old male, has two children, likes cars, engaged in high-tech industries), and after matching with the company’s internal user information, finds users with high similarity (user B, 40-year-old male, engaged in high-tech industry), it is reasonable to speculate that B may also like cars, which in turn makes product recommendations more accurate. Similarly, product-based recommendations can also be made. For example, if a product is pushed to a target group, as long as the incoming user fits the profile of this group, the product will be recommended to him.

The server 1 calculates and obtains user intention grouping information through a deep learning model based on the first user information, user login information, web page browsing track information, and attention information. Next, the analysis module 11 of the server 1 provides user recommended content according to the user intention grouping information and multiple content information, and transmits the user recommended content to the terminal device CD for display.

Please refer to FIG. 9 . FIG. 9 is another schematic diagram of the content recommendation system of the present invention providing user recommended content according to user intention grouping information.

The content recommendation system in FIG. 9 is similar to the content recommendation system in FIG. 7 , and the server 1 can also add user behavior analysis information to further increase the diversity of user recommended content.

Server 1 will calculate and obtain user intention grouping information through a deep learning model based on user login information, web browsing track information, and attention information. Next, the analysis module 11 of the server 1 provides user recommended content according to the user behavior analysis information, user intention grouping information and multiple content information, and transmits the user recommended content to the terminal device CD for display. User behavior analysis information is the information obtained based on the analysis of multiple user behaviors, for example: products or services that are more frequently purchased among multiple content information in the web group WPG, or when users are interested in specific types of When the product or service has a high purchase intention, the product or service has a higher probability of successful transaction. That is, the user behavior analysis information is content information after user behavior analysis. The analysis module 11 of the server 1 will add these products or services to the user's recommended content and provide them to the user together.

Please refer to FIG. 10 . FIG. 10 is another schematic diagram of the content recommendation system of the present invention providing user recommended content according to user intention grouping information.

The content recommendation system in FIG. 10 is similar to the content recommendation system in FIG. 7. The server 1 can add the user’s first user information and user behavior analysis information at the same time, so as to further strengthen the analysis accuracy of user intention grouping information and Diversity of user recommended content.

In this embodiment, the server 1 calculates and obtains user intention grouping information through a deep learning model according to the first user information, user login information, web browsing track information, and attention information. Next, the analysis module 11 of the server 1 provides user recommended content according to the user intention grouping information and multiple content information, and transmits the user recommended content to the terminal device CD for display.

The analysis module 11 of the server provides user recommendation content according to the user behavior analysis information, user intention grouping information and multiple content information, and transmits the user recommendation content to the terminal device CD for display.

User behavior analysis information is content information after user behavior analysis. The analysis module 11 of the server 1 will add the products or services related to the user behavior analysis information into the user recommended content and provide them to the user.

[Second embodiment]

Please refer to FIG. 11 . FIG. 11 is a flowchart of a content recommendation method according to a second embodiment of the present invention. The content recommendation method in this embodiment is applicable to the content recommendation system SYS1 mentioned above, and its structure and functions will not be repeated here.

The content recommendation method in this embodiment includes the following steps:

providing a webpage group (step S110);

Receive a plurality of user intention information (step S120);

According to the plurality of user information and a deep learning model, a user intention grouping information is calculated (step S130);

According to the user intention grouping information, a recommended content is provided (step S140 ).

In step S110 , the content recommendation system SYS1 includes a server 1 . The server 1 can receive a plurality of user intention information of the terminal device CD. The terminal device CD includes at least one display module (not shown in the figure) for displaying a webpage group WPG. The webpage group WPG includes a plurality of webpage information P1-P10. Each web page information P1-P10 is at least connected to another web page information. In addition, each webpage information P1-P10 includes a plurality of area information. That is, each web page information P1-P10 can be divided into different areas, for example: a first area, a second area and a third area. The first area, the second area and the third area are sequentially arranged. The display module (not shown) of the terminal device CD can determine which area of the webpage information P1-P10 it is according to the currently displayed area.

Furthermore, the webpage browsing track information can also consider the different regions where the user viewed the webpage information P1-P10, for example, the multiple regions information P11-P13 of the webpage information P1. For example, start viewing from the first area information P11, then view the third area information P13, and then view the second area information P12. In other embodiments, a gaze tracking technology of the terminal device CD may be used to track the user's gaze, so as to more accurately obtain the user's browsing track in the webpage group WPG.

Please refer to FIG. 2 , the webpage group WPG includes ten webpage information P1-P10. The webpage information P1 is the initial webpage of the website, and the other webpage information P2, webpage information P5, and webpage information P8 are links to the webpage information P1. That is, each web page information will link at least one other web page information. In this embodiment, in addition to the basic information, initial recommended content can be set for the user to choose.

In step S120, the user intention information at least includes web browsing track information and attention degree information. Wherein, the webpage browsing trace information includes a browsing trace of a plurality of webpage information P1-P10. The attention degree information at least includes a dwell time of each web page information P1-P10.

The webpage browsing track information is the track of the user viewing the plurality of webpage information P1-P10 in the webpage group WPG. For example, the user starts to browse from the webpage information P1 first, then clicks to enter the webpage information P2, then enters the webpage information P3, and then enters the webpage information P4. At this time, the user's webpage browsing track information WP1 is: "webpage information P1, webpage information P2, webpage information P3, webpage information P4". Other webpage browsing track information WP2 can be "webpage information P1, webpage information P8, webpage information P9, webpage information P10".

The attention degree information at least includes the user's stay time on each web page information P1-P10. For example, stay on the webpage information P1 for 30 seconds to browse. Stay on page information P2 for 1 minute to browse. Stay on the webpage information P3 and the webpage information P4 for 5 minutes respectively to browse. In addition, the attention degree information may also include user's click information on each webpage information P1-P10. For example, click on the webpage information P3.

In step S130, the server 1 calculates user intention grouping information through a deep learning model based on the user's web browsing track information and attention degree information on each web page information P1-P10 of the web page group WPG. The server 1 then provides user recommended content according to user intention grouping information and a plurality of content information.

In this embodiment, a necessary information and a non-essential information are used to perform the calculation of the deep learning model, so as to obtain the aforementioned attention degree information. For example: Recurrent neural network (RNN), long short-term memory model (Long Short-Term Memory, LSTM) in time recurrent network model, or bidirectional long short-term memory model (Bidirectional Long Short-Term Memory, BiLSTM).

As shown in FIG. 3 , the necessary information includes: the page type of the webpage information P1-P10, the stay time or viewing date of each webpage information P1-P10. Non-essential information includes: gender, company, country and other information in user information. In FIG. 3 , the attention degree information of each web page information P1-P10 is calculated by using a cycle network model. In addition, in this embodiment, the page types of the web page information P1-P10 can be preset with different weights for the calculation of the deep learning model.

As shown in Figure 4, the full-connection layer deep learning model is calculated using web browsing track information and attention information to calculate user intention grouping information. The full-connection layer deep learning model includes an input layer and multiple hidden computing layers. and the output layer. The input layer is web browsing track information and attention. The output layer is user intention grouping information. Wherein, the user intention grouping information may include: a low-level intention information, a middle-level intention information and a high-level intention information. Low-level intention information is: only browsing the website, without obvious purchase intention. Intermediate intention information is: there is an intention to purchase a product or service, but the purchase item is not clear. The high-level intention information is: there is a definite intention to purchase a predetermined product or predetermined service.

In step S140, the terminal device CD will firstly display the web page group WPG. As mentioned earlier, the webpage group WPG will firstly display initial recommended content to provide user reference. The terminal device CD will send the user information, the webpage browsing track information and attention degree information of the user browsing webpage group WPG to the server 1 . In this embodiment, the terminal device CD includes a terminal communication module (not shown). The terminal communication module (not shown in the figure) can be a wired communication module or a wireless communication module.

The server 1 includes an analysis module 11 , a storage module 12 and a communication module 13 . The analysis module 11 is electrically connected to the storage module 12 and the communication module 13 .

The server 1 receives the user information through the communication module 13, the webpage browsing track information of the user browsing webpage group WPG, and the attention degree information are sent to the server 1, and the server 1 will further use the analysis module 11 to analyze the user information. , and calculate the webpage browsing track information and attention degree information of the user browsing webpage group WPG. That is, the analysis module 11 of the server 1 will use user information and web browsing track information, and use a deep learning model to calculate attention degree information.

At this time, the terminal device CD will firstly send the user login information to the server 1 . That is, the terminal device CD will send the information of whether the user has logged into the webpage group WPG to the server 1, for example: logged in (new member), logged in (old member), not logged in.

Furthermore, the analysis module 11 of the server 1 uses the web page browsing track information and attention degree information to calculate user intention grouping information using a type of neural model. The analysis module 11 of the server 1 will further provide a user recommended content to the terminal device CD for display according to the user intention grouping information and multiple content information. The content information may include product information or service information.

When the user sees the displayed user recommended content, the web page group WPG of the terminal device CD can send a user preference information to the server according to whether the user clicks on the user recommended content or clicks on the dislike user recommended content 1. The user preference information is information about the user's preference or clicks on the user-recommended content. The analysis module 11 of the server 1 will further calculate the user login information, user preference information, web browsing track information and attention degree information through deep learning models such as deep learning, and calculate the user intention grouping information again. Then, the server 1 will provide another recommended content for the user according to the recalculated user intention grouping information and multiple content information. The analysis module 11 of the server 1 will use the communication module 13 to transmit the re-provided user recommendation content to the terminal device CD for display, and provide the calculated user recommendation content to the user again.

The analysis module 11 of the server 1 can calculate the content recommended by the user every predetermined time interval. In each predetermined time interval, the user's web browsing track information and attention degree information will change, so the user's recommended content will also be adjusted accordingly. The predetermined time interval can be adjusted according to actual needs, which is not limited in the present invention.

In addition, the server 1 stores the user login information, user preference information, web page browsing track information, attention degree information, etc. corresponding to the user in the storage module 12 . That is, to establish a database file for users as the basis for improving user experience and providing better services.

In addition, the server 1 can also collect the first user information of the user to further enhance the analysis accuracy of the user intention grouping information. The first user information is various identity information provided by the user in the webpage group WPG, such as photos, personal information, and the like. In addition, the first user information may also include information of other cooperative websites or cooperative organizations and information without privacy concerns in other databases.

That is, the server 1 will calculate and obtain user intention grouping information through a deep learning model according to the first user information, user login information, web page browsing track information, and attention information. Next, the analysis module 11 of the server 1 provides user recommended content according to the user intention grouping information and multiple content information, and transmits the user recommended content to the terminal device CD for display.

Furthermore, the server 1 can also add user behavior analysis information to further increase the richness of user recommended content. User behavior analysis information is the content information obtained by analyzing multiple user behaviors, for example: products or services that are more frequently purchased among the multiple content information in the web group WPG, or when users are interested in specific Products or services that have a higher probability of successful transaction when the type of product or service has a higher purchase intention. That is, the user behavior analysis information is content information after user behavior analysis. The analysis module 11 of the server 1 will add these products or services to the user's recommended content and provide them to the user together.

That is, the server 1 will calculate and obtain user intention grouping information through a deep learning model based on user login information, web browsing track information, and attention information. Next, the analysis module 11 of the server 1 provides user recommended content according to the user behavior analysis information, user intention grouping information and multiple content information, and transmits the user recommended content to the terminal device CD for display. That is, the analysis module 11 of the server 1 will add the products or services related to the user behavior analysis information into the user recommendation content and provide them to the user together.

Furthermore, the server 1 can also add the first user information of the user at the same time to further strengthen the analysis accuracy of the user intention grouping information. In addition, the analysis module 11 of the server will also provide user recommended content according to the user behavior analysis information, user intention grouping information and multiple content information, and transmit the user recommended content to the terminal device CD for display.

User behavior analysis information is content information after user behavior analysis. The analysis module 11 of the server 1 will add the products or services related to the user behavior analysis information into the user recommended content and provide them to the user.

At this time, the server 1 calculates and obtains user intention grouping information through a deep learning model according to the first user information, user behavior analysis information, user login information, web browsing track information, and attention degree information.

[Advantageous Effects of Embodiment]

One of the beneficial effects of the present invention is that the content recommendation system and content recommendation method provided by the present invention can effectively use the user's web browsing track information and attention information in the web page group to accurately provide users with appropriate recommended content , which can increase the user's high interest in the website group and related user experience.

The content disclosed above is only a preferred feasible embodiment of the present invention, and does not therefore limit the scope of the patent application of the present invention. Therefore, all equivalent technical changes made by using the description and drawings of the present invention are included in the application of the present invention. within the scope of the patent.

SYS1: Content Recommendation System 1: Server CD: terminal device S110-S140: Steps WP1, WP2: Browse track P1-P10: Web page information P11: Information of the first region P12: Second area information P13: Third Region Information 11: Analysis module 12: Storage module 13: Communication module WPG: Web Page Group

FIG. 1 is a schematic diagram of a content recommendation system according to a first embodiment of the present invention.

FIG. 2 is a schematic diagram of a web page group in the present invention.

FIG. 3 is a schematic diagram of calculating attention degree information by the content recommendation system of the present invention using a deep learning model.

FIG. 4 is a schematic diagram of calculating user intention grouping information by the content recommendation system of the present invention using a deep learning model.

FIG. 5 is another schematic diagram of the content recommendation system of the present invention using a deep learning model to calculate user intention grouping information.

Fig. 6 is a functional block diagram of the content recommendation system of the present invention.

FIG. 7 is a schematic diagram of the content recommendation system of the present invention providing user recommended content according to user intention grouping information.

FIG. 8 is another schematic diagram of the content recommendation system of the present invention providing user recommended content according to user intention grouping information.

FIG. 9 is another schematic diagram of the content recommendation system of the present invention providing user recommended content according to user intention grouping information.

FIG. 10 is another schematic diagram of the content recommendation system of the present invention providing user recommended content according to user intention grouping information.

Fig. 11 is a flow chart of the content recommendation method according to the second embodiment of the present invention.

SYS1: Content Recommendation System

1: Server

CD: terminal device

Claims (7)

A content recommendation system, comprising: a server that receives a plurality of user intention information of a user of a terminal device, wherein the user intention information at least includes webpage browsing track information and attention degree information, and the webpage Browsing trajectory information includes a browsing sequence trajectory of multiple webpage information, and the attention degree information includes at least a dwell time and a number of webpage browsing times for each of the webpage information; the terminal device includes a display module for displaying a webpage group, the webpage group includes a plurality of the webpage information, each of the webpage information is at least connected to another webpage information, and each of the webpage information corresponds to a predetermined weight; the server according to The webpage browsing track information and the attention degree information are calculated through a deep learning model to obtain a user intention grouping information; the user intention grouping information includes a low-level intention information, a middle-level intention information and a high-level intention information information, the low-level intention information is only browsing the website without obvious purchase intention, the middle-level intention information is the intention to purchase products or services, but the purchase items are not clear, and the high-level intention information is the clear purchase intention information The intention of pre-ordering products or pre-ordering services; the server provides a user recommended content according to the user intention grouping information and a plurality of content information, and the plurality of content information is set in the webpage group; The display module displays the user recommended content and sends a user preference information to the server according to whether the user likes the user recommended content; the server according to a user login information, the user Preference information, the web browsing track information and the attention information, through the A deep learning model is used to calculate a second user intention grouping information; and the server provides a second user recommendation content according to the second user intention grouping information. The content recommendation system as described in claim 1, wherein the server calculates the usage information through the deep learning model according to the first user information, the web browsing track information, and the attention degree information. The user's intention group information, the server provides the user recommended content according to the user intention group information and the plurality of content information, the first user information is the user's information in the webpage group The user login information in the . The content recommendation system as described in claim 2, wherein the server provides the user recommended content according to a user behavior analysis information, the user intention grouping information and the plurality of content information, and the use User behavior analysis information is content information after user behavior analysis. A content recommendation method, comprising: providing a webpage group by a terminal device, wherein the webpage group includes a plurality of webpage information, each of the webpage information is at least connected to another webpage information; a server Receive a plurality of user intention information of a user, each of the user intention information includes at least a webpage browsing track information and a attention degree information, and the webpage browsing track information includes a browsing track of the plurality of webpage information , the attention information at least includes a dwell time and a number of page views of each of the webpage information; the server calculates a user intention according to the plurality of user intention information through a deep learning model Grouping information; the user intention grouping information includes a low-level intention information, a middle-level intention information and a high-level intention information. Information is available for purchase or service intentions, but the purchase items are not clear, the high-level intention information is the intention to purchase predetermined products or predetermined services; the server provides a User recommended content, the plurality of content information is set in the webpage group; a display module of the terminal device displays the user recommended content and sends a message according to whether the user likes the user recommended content The user preference information is sent to the server; the server calculates and obtains it through the deep learning model according to a user login information, the user preference information, the web page browsing track information and the attention degree information A second user intention grouping information; and the server provides a second user recommended content according to the second user intention grouping information. The content recommendation method as described in claim 4, wherein, the server calculates the usage based on the first user information, the webpage browsing track information, and the attention degree information through the deep learning model. The user's intention group information, the server provides the user recommended content according to the user intention group information and the plurality of content information, the first user information is the user's information in the webpage group The user login information in the . The content recommendation method as described in claim 5, wherein the server provides the user recommended content according to a user behavior analysis information, the user intention grouping information and the plurality of content information, and the user Behavior analysis information is content information after user behavior analysis. The content recommendation method as described in claim 4, wherein the user intention information is obtained by tracking the user's line of sight to obtain the web browsing track information and the attention degree information.

Images