TWI630571B - Essay recommendation method and a computer-readable media - Google Patents

Abstract

The article recommendation method is used to recommend an article. Firstly, an influence network is built. The influence network has a plurality of primary nodes and a plurality of child nodes, and each of the child nodes has different probability values respectively affected by the primary node. Then, the influence scores of the master nodes are calculated according to different probability values, and the master node with the highest influence score is set as the first master node. Then reconstruct the influence network to exclude the influence of the first master node on other master nodes and child nodes, and calculate the influence scores of the remaining master nodes after excluding the first master node, and the master node with the highest influence score among them Set to the second master node. The first master node and the second master node are objects that recommend the article.

Description

Article recommendation method and computer readable media

The present invention relates to a recommended method, and in particular to an article recommendation method.

In the process of message transmission, some people are more likely to spread this message widely and even become more popular than others. Such people are called "influential people."

Traditionally, the advertising recommendation system saves the cost of message dissemination by sending trials or gifts to these influential people for trial use and passing the trial results or product recommendation messages. Therefore, how to maximize this communication effect, that is, how to effectively identify who is the most influential person, becomes the most critical issue for effective message dissemination.

A technical aspect of the present invention is to provide an article recommendation method for recommending an article, comprising the steps of: constructing an influence network having a plurality of master nodes and a plurality of child nodes, Each child node has different probability values respectively affected by the master nodes; according to the different probability values, the influence scores of the master nodes are respectively calculated; and the master nodes with the highest influence scores in the master nodes are recorded as a first master node; rebuilding the influence network, further comprising: excluding, in the influence network, the influence of the first master node on other master nodes and the child nodes, wherein each child node has After the first primary node, different new probability values respectively affected by other primary nodes; according to the different new probability values, respectively calculating the influence scores of the remaining primary nodes after excluding the first primary node; and recording the first primary The master node with the highest influence score among the remaining master nodes outside the node serves as a second master node; wherein the first master node and the second master node are objects that recommend the article.

In summary, the technical solution of the present invention has obvious advantages and beneficial effects compared with the prior art. Through the above technical solution, the user with the greatest influence can be searched in an influence network, and the article can be sent for the object, and at the same time, the "active group" can be avoided, thereby maximizing the advertising benefit.

The above description will be described in detail in the following embodiments, and further explanation of the technical solutions of the present invention will be provided.

100 and 600‧‧‧ article recommendation methods

200‧‧‧Influence Network

101-105, 601-605‧‧‧ steps

A, B, C‧‧‧ primary node

D, E, F, G, H, I, J, K, L, M, N, O‧‧‧ child nodes

The above and other objects, features, advantages and embodiments of the present invention will become more <RTIgt; <RTIgt; </ RTI> <RTIgt; </ RTI> <RTIgt; </ RTI> <RTIgt;

Figure 2 is a schematic diagram of an influential network constructed in accordance with an embodiment of the present invention.

Figure 3A shows the influence path map corresponding to the master node A.

Figure 3B shows the influence path map corresponding to the master node B.

Figure 3C shows the influence path map corresponding to the master node C.

Figure 4A shows the probability distribution diagram of the recommended articles received by each hierarchical node when A is the main node.

Figure 4B shows the probability distribution diagram of the recommended articles for each hierarchical node when B is the main node.

Figure 4C shows the probability distribution diagram of the recommended articles received by each hierarchical node when C is the main node.

In FIG. 5A, after the main node A is excluded, when B is the main node, the probability distribution map of the recommended article is received by each hierarchical node.

In the fifth section, after the main node A is excluded, when C is the main node, the probability distribution map of the recommended articles is received by each hierarchical node.

Figure 6 is a flow chart showing an article recommendation method according to another embodiment of the present invention.

Figure 7A shows the probability and correlation depth distribution of each class node receiving the recommended article when A is the main node.

Figure 7B shows the probability and correlation depth distribution of each class node receiving the recommended article when B is the main node.

Figure 7C shows the probability and correlation depth profile of each class node receiving the recommended article when C is the main node.

Figure 8 shows the main node A, when B is the main node, each The hierarchical node receives the probability of the recommended article and the associated depth profile.

In order to make the description of the present invention more detailed and complete, reference can be made. The accompanying drawings and the various embodiments set forth below are in the However, the embodiments provided are not intended to limit the scope of the invention, and the description of the operation of the structure is not intended to limit the order of its execution, and any device that is recombined by the components produces equal devices. The scope covered by the invention.

The drawings are for illustrative purposes only and are not drawn to the original dimensions. On the other hand, well-known elements and steps are not described in the embodiments to avoid unnecessarily limiting the invention.

The method for recommending the article of the present invention is searched by prior to the human network. The most influential user, and send the article to the user, use the influence of the user to expand the influence of the article, and use the user's network to send and recommend articles, thereby reducing the transmission cost. And maximize the benefits of communication.

FIG. 1 is a diagram showing an article recommendation method according to an embodiment of the present invention. Flow chart. The article recommendation method 100 can be implemented as a computer program and stored in a computer readable recording medium, so that a computer or an electronic device can read the recording medium and execute the method of playing multimedia on the virtual desktop. Computer-readable recording media can be read-only memory, flash memory, floppy disk, hard disk, optical disk, flash drive, tape, network accessible database or familiar with the art can easily think of the same Functional computer readable Record media.

This article recommends method 100, first establishing an impact in step 101. Power network. In a network, when a user sends any kind of interaction to another user, a directional probability link can be established between them based on the importance of the interaction. The second interaction creates another directional probabilistic link, which in turn creates an impact network. In one embodiment, the network of influences can be built on a social networking site, such as Facebook. 2 is a schematic diagram of an influence network constructed according to an embodiment of the present invention. In this embodiment, the influence network 200 includes a plurality of nodes, and each node can represent a user and a node. The connection relationship with other nodes can represent the interpersonal context of the node and its influence.

Next, please refer to both Figure 1 and Figure 2, when this influence network After the road 200 is completed, in step 102, the influence breadth score of each major node in the influence network is calculated, and in step 103, according to the calculated influence score of the main node influence, the largest influence spread score is selected. The master node, as the node with the greatest impact breadth, is recorded in the node selection list as the object of the article. In the influence network 200, the nodes affected by the nodes A, B, and C are the most connected according to the respective connection relationships of the nodes, so the nodes A, B, and C can be selected as the main ones of the influence network 200. Nodes, and calculate the influence breadth scores of the master nodes A, B, and C. However, the impact breadth score calculation can also be performed for each node. Among them, FIG. 3A shows an influence path map corresponding to the main node A. Figure 3B shows the influence path map corresponding to the master node B. Figure 3C shows The influence path map corresponding to the master node C. The child node directly coupled to the primary node is a first hierarchical node, the child node coupled to the first hierarchical node is a second hierarchical node, and so on. In an embodiment, when the master node receives an article, the master node has half, or 50%, of the chance to re-recommended the received article to the node of the first hierarchy, so the first hierarchical node receives The probability of this article is 50%. The first level node also has a 50% chance of recommending the received article to the second level node, so the probability of the second level node receiving the article is 25%. The second-level node also has a 50% chance of recommending the received article to the third-level node, so the probability of the third-level node receiving the article is 12.5%, and so on, and the nodes at each level can be calculated. The chance to recommend an article. Among them, Figure 4A shows the probability distribution map of the recommended articles received by each hierarchical node when A is the main node, and then the probability of receiving the recommended articles by each hierarchical node is added, and the article can be obtained from the primary node A. The spread breadth of the spread to all levels of nodes is 6.218755. Figure 4B shows the probability distribution diagram of the recommended articles received by each hierarchical node when B is the main node. According to this, the probability that each hierarchical node receives the recommended article can be obtained, and the article can be spread from the primary node B to The influence breadth score of each hierarchical node is 4.8125. Figure 4C shows the probability distribution graph of the recommended articles received by each hierarchical node when C is the main node. According to this, the probability that each hierarchical node receives the recommended article is obtained, and the article can be diffused from the primary node C to The influence breadth score of each hierarchical node is 5.1171875. Among the three master nodes, the influence score of the master node A is the highest, so in step 103, the master node A is selected as the node with the greatest influence breadth, and is recorded in the node selection list as the object to be sent.

However, there are so-called "active groups" in the social network. In the "active groups", the information exchange speed is very fast, that is, the messages received by the nodes in the "active group" will be quickly spread. Going out, the nodes with high impact breadth scores often appear in these "active groups". If you simply select the nodes in this "active group", the information distributed according to this will be restricted to this fixed group. It is difficult to pass the message to a place other than this group. Therefore, in order to avoid being restricted to the "active group", in step 104, an influence network is reconstructed again according to the selected master node A, and in this reconstruction of the influence network, the node with the greatest influence breadth is selected again. And recorded in the node selection list, as the sending object of the article as the article sending object.

In the step of reconstructing the influence network, first in step 1041, the probability that other nodes are affected by other master nodes after excluding the selected master node is calculated. Next, in step 1042, the influence breadth score of each main node in the reconstructed influence network is calculated according to the probability of each node, and in step 1043, according to the calculated influence score of the main node influence, the highest influence spread score is selected. The master node, which is the node with the greatest influence breadth in the influence network after this reconstruction, is recorded in the node selection list. In order to avoid being restricted to "active groups", in the case of rebuilding the influence network, the case will first exclude the influence of the master node A on each child node, and then calculate the child nodes based on this, and only accept the master node B or the master node C. The probability of influence, and the breadth score of the master node B and the master node C are calculated accordingly.

In FIG. 5A, after the main node A is excluded, when B is the main node, the probability distribution map of the recommended article is received by each hierarchical node. Where area 501 The child node is only coupled to the master node A, that is, only affected by the master node A. Since the master node A is excluded, the probability of the child node at the region 501 receiving the recommended article is zero. According to FIG. 4A, the probability that the master node B receives the recommended article of the master node A is 25%. Therefore, after the master node A is excluded, the probability of receiving the article recommended by the non-master node A is 75%. Sub-node D will receive the recommended articles of the master node A and the master node B at the same time, and according to the 4A map, the probability that the child node D receives the recommendation article of the master node A is 50%, so after the master node A is excluded, it is received. The probability that the non-master node A recommends the article is 50%, and the master node B has a 50% chance to recommend the received article to the child node D, so the probability that the child node D receives the recommended article is 50%*50%. =25%. The child node H only receives the article recommended by the child node D, and the child node D has a 50% chance of recommending the received article to the child node H, so the probability of the child node H receiving the recommended article is 25%* 50% = 12.5%. In addition, the child nodes E, F, G only receive the article recommended by the master node B, and the master node B has a 50% chance of recommending the received article to the child nodes E, F, G, and thus the child node E, F, G The probability of receiving a recommended article is 75%*50%=37.5%, and the probability that the remaining child nodes receive the recommended article can be deduced by analogy. After the probability that each node receives the recommended article is established, the probability that each node receives the recommended article can be added according to this, and after the main node is excluded, the influence spread of the article from the primary node B to each node is 2.53125.

On the other hand, in the 51Bth, when the main node A is excluded, when C is the main node, the probability distribution map of the recommended article is received by each hierarchical node. The node at the area 502 is only coupled to the primary node A, that is, only affected by the primary node A. Since the primary node A is excluded, the subsection at the region 502 The probability of receiving a recommended article is zero. According to FIG. 4A, the probability that the master node C receives the recommended article of the master node A is 25%. Therefore, after the master node A is excluded, the probability of receiving the article recommended by the non-master node A is 75%. The child nodes I, J, K, L, M, N, O will receive the recommended articles of the master node A and the master node C according to the 4A diagram, and the child nodes I, J, K, L, M, N, O are received. The probability that the main node A recommends the article is 50%, so after excluding the master node A, the probability of receiving the non-master node A recommendation article is 50%, and the master node C has a 50% chance to recommend the article received again. For the child nodes I, J, K, L, M, N, O, the probability that the child nodes I, J, K, L, M, N, O receive the recommended article is 50% * 50% = 25%. After the probability that each node receives the recommended article is established, the probability that each node receives the recommended article is added, and after the main node is excluded, the influence spread of the article from the primary node C to each node is 2.5. Therefore, in step 1043, the node having the greatest influence breadth of the influence network after the reconstruction of the master node B is selected, and recorded in the node selection list as the object to be sent.

Then, in step 105, it is determined whether the number of nodes selected in the node selection list reaches the set number. If the condition is reached, the node selection is terminated, and the article is sent according to the selected node, otherwise step 104 is performed again, and the reconstruction effect is performed. Force the network to exclude this choice from the master node to influence the impact of the network. In another embodiment, it is also possible to determine whether the number of nodes that can be affected by the selected node in the node selection list has covered a certain proportion of the node to be affected. If the condition has been reached, the node is selected and the node is selected according to the selected node. , send the article. In an embodiment, the above two conditions may also be determined at the same time, and as long as one of the conditions is reached, the node is terminated. Select and send the article according to the selected node.

In addition to evaluating the breadth of influence, in another embodiment, the depth of influence is evaluated. Figure 6 is a flow chart showing an article recommendation method according to another embodiment of the present invention. The process is similar to that of FIG. 1. Similarly, the article recommendation method 600 can be implemented as a computer program and stored in a computer readable recording medium, so that a computer or an electronic device can read the recording medium. After performing the method of playing multimedia on the virtual desktop. Computer-readable recording media can be read-only memory, flash memory, floppy disk, hard disk, optical disk, flash drive, tape, network accessible database or familiar with the art can easily think of the same The function of the computer can read the recording medium.

This article recommends method 600. First, in step 601, an overview of the influence network as shown in FIG. 2 is constructed. In one embodiment, the network of influences can be built on a social networking site, such as Facebook. The influence network 200 includes a plurality of nodes, each node can represent a user, and the connection relationship between a node and other nodes can represent the interpersonal context of the node and its influence.

Then, after the impact network 200 is completed, the impact depth score of each major node in the influence network may be calculated in step 602, and in step 603, according to the calculated influence score of the master node. The master node in which the depth score is the largest is selected as the node with the greatest influence depth, and is recorded in the node selection list as the object to be sent. In this influence network 200, the nodes A, B, and C are the main nodes in the influence network 200 as described above. Figure 3A shows the corresponding node A. The impact path map. Figure 3B shows the influence path map corresponding to the master node B. Figure 3C shows the influence path map corresponding to the master node C. Wherein, when the master node receives an article, half of the master node, that is, 50%, has the chance to recommend the received article to the first class node, so the probability of the first class node receiving the article is 50%. In addition, for the master node, the article is recommended to the next layer of child nodes, that is, the first level node, which represents the master node has a deep influence, so the depth influence can be weighted, and the corresponding probability is 50%*2. On the other hand, the first hierarchical node also has a 50% chance of recommending the received article to the second hierarchical node, so the probability of the second hierarchical node receiving the article is 25%. For the master node, the article is recommended to the second hierarchical node, and the corresponding probability score after weighting the depth influence is 25%*3. The second level node also has a 50% chance of recommending the received article to the third level node, so the probability of the third level node receiving the article is 12.5%. For the master node, the article is recommended to the third-level node, and the corresponding probability score after weighting the depth influence is 12.5%*4. Among them, Figure 7A shows the probability and related depth distribution of each class node receiving the recommended article when A is the main node. Figure 7B shows the probability and correlation depth distribution of each class node receiving the recommended article when B is the main node. Figure 7C shows the probability and correlation depth profile of each class node receiving the recommended article when C is the main node. The depth impact scores for each node can then be calculated. In an embodiment, the following formula calculates the depth probability of each hierarchical node corresponding to the primary node: depth probability = probability (n) * (1.0 - product (1.0 - depth distribution (u) / probability (n))), wherein u is the branch of n.

Taking the 7B picture as an example, the depth probability of the primary node is 0, the probability of the depth of the first hierarchical node corresponding to the primary node is 0.0625, and the probability of the second hierarchical node corresponding to the primary node is 0.40625, and the probability of the third hierarchical node is The depth machine rate corresponding to the master node is 0.189209, the depth factor of the fourth-level node probability corresponding to the master node is 0.279541, and the depth factor of the fifth-level node probability corresponding to the master node is 0.0625. Therefore, if the depth expectation value is considered, at this time, the master The depth expectation of Node B, that is, the depth score is: depth expectation (depth score) = Depth (d) * depth probability where d is the depth and n is the deepest depth. Therefore, the depth expectation is (1*0.0625)+(2*0.40625)+(3*0.189209)+(4*0.279541)+(5*0.0625)=2.873291

If the node with the highest influence depth score selected in this step is A, similarly, in order to avoid being restricted to the "active group", in step 604, an influence network is reconstructed according to the selected master node A again. In this reconstruction of the influence network, the node with the greatest influence depth is selected again and recorded in the node selection list as the object to be sent as the article transmission object.

In the step of reconstructing the influence network, first in step 6041, the probability that other nodes are affected by other master nodes after excluding the selected master node is calculated. Next, in step 6042, the influence depth scores of the main nodes in the reconstructed influence network are calculated according to the probability of each node, and in step 6043, the shadows are selected according to the calculated influence scores of the master nodes. The master node with the highest depth score is the node with the greatest influence depth in the influence network after this reconstruction, and is recorded in the node selection list. Similarly, in the case of Figure 7B, in order to avoid being restricted to "active groups", in the case of rebuilding the influence network, the case will first eliminate the influence of the master node A on each node, and then calculate the node based on this, only accept the master. The probability that Node B affects, and the depth score of Master Node B is calculated accordingly.

Figure 8 shows the main node A, when B is the main node, Each level node receives the probability of the recommended article and the relevant depth profile. The node at the area 801 is only coupled to the master node A, that is, only affected by the master node A. Since the master node A is excluded, the probability of the node at the region 801 receiving the recommended article is zero. According to Fig. 8, the probability that the master node B receives the recommended article of the master node A is 25%. Therefore, after the master node A is excluded, the probability of receiving the article recommended by the non-master node A is 75%. The first hierarchical node D of the primary node B receives the recommended articles of the primary node A and the primary node B at the same time, and according to the eighth figure, the probability that the node D receives the recommended article of the primary node A is 50%, so the primary node is excluded. After A, the probability of receiving a non-master node A recommendation article is 50%, and the master node B has a 50% chance to recommend the received article to the first class node D, so node D receives the recommendation of the master node B. The probability of the article is 50%*50%=25%. In addition, for the master node B, the article is recommended to the first hierarchical node D, which represents the deep influence of the primary node B, so the depth influence can be weighted, corresponding to The chance is 25%*2. Node H is the second-level node, only receives the article recommended by node D, and node D has a 50% chance to recommend the received article to node H, so the probability of node H receiving the recommended article is 25%. *50%=12.5%. Plus If the weight of the master node B is influential, the corresponding probability is 12.5%*3. In addition, the nodes E, F, and G are the first hierarchical nodes, and only the articles recommended by the primary node B are received, and the primary node B has a 50% chance of recommending the received articles to the nodes E, F, and G, thus The probability that the nodes E, F, and G receive the recommended article of the primary node B is 75%*50%=37.5%, and the weight of the primary node B is further weighted, and the corresponding probability is 37.5%*2. The probability that the remaining nodes receive the recommended article can be deduced by analogy. After each node receives the recommendation article and the probability of weighting the depth influence is established, the depth probability of each hierarchical node probability corresponding to the primary node B can be calculated according to the following formula: depth probability (n)=probability (n)*(1.0- Product (1.0-depth distribution (u) / probability (n))), where u is the branch of n.

Taking Figure 8 as an example, the depth probability of the master node is 0.25, the probability of the depth of the first-level node corresponding to the master node is 0.0625, and the probability of the depth of the second-level node corresponding to the master node is 0.40625, and the probability of the third-level node is The depth probability corresponding to the master node is 0.1875, and the depth factor of the fourth-level node probability corresponding to the master node is 0.09375. Therefore, if the depth expectation value is considered, the depth expectation value of the master node B, that is, the depth score is (1*). 0.0625) + (2 * 0.40625) + (3 * 0.1875) + (4 * 0.09375) = 1.8125. If the primary node B has the maximum impact depth score, then in step 6043, the node having the greatest influence depth in the influence network after the primary node B is reconstructed is selected and recorded in the node selection list as the object to be sent.

Then, in step 605, it is determined whether the number of nodes selected in the node selection list reaches the set number. If the condition is reached, the node selection is terminated, and the article is sent according to the selected node, otherwise the re-execution is performed. In step 604, the influence network is reconstructed, and the influence of the master node on the influence network is excluded. In another embodiment, it is also possible to determine whether the number of nodes that can be affected by the selected node in the node selection list has covered a certain proportion of the node to be affected. If the condition has been reached, the node is selected and the node is selected according to the selected node. , send the article. In an embodiment, the two conditions may be determined at the same time. As long as one of the conditions is met, the node selection is terminated, and the article is sent according to the selected node.

According to the method of the present invention, the user with the greatest influence can be searched in an influence network, and the article can be sent for the object, and at the same time, the "active group" can be avoided at the same time, so Advertising effectiveness.

The present invention has been disclosed in the above embodiments, and is not intended to limit the invention, and it is intended that various modifications and changes may be made without departing from the spirit and scope of the invention. The scope of the invention is defined by the scope of the appended claims.

Claims (10)

An article recommendation method for recommending an article includes the following steps: constructing an influence network having a plurality of master nodes and a plurality of child nodes, each child node having a respective master node Different probability values of the influence; according to the different probability values, respectively calculate the influence scores of the master nodes; record the master nodes with the highest influence scores among the master nodes as the first master node; reconstruct the influence network The method further includes: excluding, in the influence network, the influence of the first primary node on the other primary nodes and the child nodes, wherein each child node has a different primary node after excluding the first primary node Different new probability values of the impact; according to the different new probability values, respectively calculate the influence scores of the remaining primary nodes after excluding the first primary node; and record the highest influence scores of the remaining primary nodes other than the first primary node The master node serves as a second master node; and transmits the article according to the first master node and the second master node. The article recommendation method of claim 1, further comprising: determining whether the number of primary nodes of the record has reached a set value; After the recorded number of master nodes reaches the set value, the article is transmitted to the master node of the records; and when the number of recorded master nodes does not reach the set value, the influence network is reconstructed according to the recorded master node. The article recommendation method of claim 1, further comprising: determining whether the number of child nodes affected by the record master node has reached a set ratio; and when the number of child nodes affected by the record master node has reached the set ratio, Transmitting the article to the master node of the records; and when the number of child nodes affected by the record master node does not reach the set ratio, the influence network is reconstructed according to the recorded master node. The article recommendation method of claim 1, wherein the influence score is an influence breadth score. The article recommendation method according to claim 4, wherein calculating the influence breadth scores of the master nodes according to the different probability values further comprises: summing the probability values of each child node as the corresponding influence of the master node Force breadth score. The article recommendation method of claim 1, wherein the influence score is an influence depth score. The article recommendation method of claim 6, wherein the probability values are probability values after weighting the sub-node depth. The article recommendation method according to claim 6 is based on the influence depth score: Where d is the depth and n is the deepest depth of the influence network. The article recommendation method described in claim 1, wherein the establishment of an influence network is performed through a personal network of a social networking website. A computer readable recording medium storing a computer program for performing an article recommendation method for recommending an article, the method comprising: constructing an influence network having a plurality of master nodes And a plurality of child nodes, each child node having different probability values respectively affected by the master nodes; according to the different probability values, calculating the influence scores of the master nodes respectively; recording the highest influence scores of the master nodes The master node, as the first master node; rebuilding the influence network, further comprising: excluding the first master node from other dominant sections in the influence network Point and the influence of the child nodes, wherein each child node has different new probability values respectively affected by the other master nodes after excluding the first master node; according to the different new probability values, respectively calculating The influence score of the remaining primary nodes after the first primary node; and the primary node having the highest impact score among the remaining primary nodes outside the first primary node, as a second primary node; and according to the first primary node and the first The second master node sends the article.