A method and device for determining label relationship based on bipartite graph relationship transfer

Abstract

This specification discloses a label relationship determination method and device based on bipartite graph relationship transfer. First, according to the relationship weights between each first label in the first label group and each second label in the second label group, the relationship transfer probabilities of each first label transferring the relationship to each other first label through different second labels can be determined for each first label. Then, according to the preset initial relationship values ​​of each other first label and the relationship transfer probabilities corresponding to the first label, the relationship value of the first label is determined. Finally, according to the relationship values ​​of each first label, the relationship between each first label is determined. By determining the relationship transfer situation between each first label, the relationship value of each first label when the relationship is transferred is obtained, and the relationship between labels with indirect relationships can be determined, which improves the recall rate of recall based on label relationships.

Description

Technical Field

[0001] The present application relates to the field of computer technology, and in particular to a method and device for determining a label relationship based on bipartite graph relationship transfer. Background Art

[0002] At present, with the development of computer technology, recommendation technology is also constantly improving. Generally, when recommending, first recall based on tags, then sort the recall results, select the top ranking results and recommend them to users. Usually, the server can extract keywords from documents through natural language processing algorithms (Natural Language Processing, NLP) such as keyword recognition, and use the extracted keywords as tags. Or, based on data mining algorithms, determine tags from massive data, etc.

[0003] The tag-based recall method in the prior art will pre-classify each tag into multiple tag groups (such as entertainment tag group, education tag group) according to the nature of the tag, and determine the relationship between the tags in each tag group. When performing tag recall, the recall is performed based on the determined relationship between the tags in each tag group.

[0004] Generally, the following method is used to determine the relationship between tags. Specifically, for each tag group, the auxiliary tag group corresponding to the tag group is first determined. Then, for each tag in the tag group, the co-occurrence rate between the tag and each tag in the auxiliary tag group is determined to obtain the relationship weight vector between the tag and the auxiliary tag group. Secondly, the relationship between the tags in the tag group is determined based on the similarity of the relationship weight vectors of the tags in the tag group.

[0005] However, the relationship between the labels in the label group depends on the co-occurrence of the labels in the label group and the labels in the auxiliary label group. Only when two labels co-occur with at least one label in the auxiliary label group can the two labels be determined to have a relationship. This makes it difficult to determine the relationship between labels with indirect relationships, resulting in a low recall rate for recall based on label relationships. Summary of the invention

[0006] The embodiments of this specification provide a method and device for determining a label relationship based on bipartite graph relationship transfer, which are used to at least partially solve the problems existing in the prior art.

[0007] This manual adopts the following technical solutions:

[0008] This specification provides a label relationship determination method based on bipartite graph relationship transfer, including:

[0009] Determine the relationship weight between each first tag in the first tag group and each second tag in the second tag group;

[0010] According to the relationship weights between the first tags and the second tags, respectively determine the first probability of the relationship transfer from the first tags to the second tags, and respectively determine the second probability of the relationship transfer from the second tags to the first tags;

[0011] For each first tag, determine, according to each first probability corresponding to the first tag and each second probability corresponding to each second tag, each relationship transfer probability of the first tag transferring the relationship to each other first tag via different second tags;

[0012] Determine the relationship value of the first tag according to the preset initial relationship values ​​of the other first tags and the relationship transmission probabilities corresponding to the first tag, wherein the higher the relationship value, the higher the probability that the first tag has an association relationship with other first tags;

[0013] According to the relationship values ​​of the first tags and the corresponding relationship transmission probabilities, the relationship values ​​of the first tags are updated until the iteration termination condition is met, and the relationship between the first tags is determined according to the relationship values ​​of the first tags.

[0014] Optionally, respectively determining the first probability of each first tag transferring the relationship to each second tag specifically includes:

[0015] For each first tag, determine the cumulative sum of the relationship weights between the first tag and each second tag as the first total weight;

[0016] First probabilities of the first tag transferring the relationship to different second tags are determined respectively according to the ratio of the relationship weights between the first tag and different second tags to the first total weight.

[0017] Optionally, respectively determining the second probability that each second tag transfers the relationship to each first tag includes:

[0018] For each second tag and each first tag, determine the cumulative sum of the relationship weights between the second tag and each other first tag as the second total weight;

[0019] Second probabilities of the second tag transferring the relationship to other first tags are determined respectively according to the ratio of the relationship weights between other first tags and the second tag to the second total weight.

[0020] Optionally, determining the probability of transferring the relationship of the first tag to other first tags via different second tags specifically includes:

[0021] Determine each sub-relationship transfer probability of the first tag corresponding to each second tag according to the product of the corresponding first probability of the first tag transferring the relationship to each second tag and the corresponding second probability of the second tag transferring the relationship to each first tag;

[0022] According to the sum of the sub-relationship transfer probabilities, the relationship transfer probabilities of the first tag transferring the relationship to other first tags via different second tags are determined.

[0023] Optionally, determining the relationship value of the first tag according to the preset initial relationship values ​​of other first tags and the relationship transfer probabilities corresponding to the first tag specifically includes:

[0024] For each other first tag, determine the sub-relationship value according to the product of the preset initial relationship value of the other first tag and the relationship transfer probability of the first tag transferring the relationship to the other first tag;

[0025] The relationship value of the first tag is determined according to the sum of the sub-relationship values ​​of the first tag.

[0026] Optionally, the sub-relationship value is determined according to the product of the preset initial relationship value of the other first tag and the relationship transfer probability of the first tag transferring the relationship to the other first tag, specifically including:

[0027] Determining first tags that conflict with each other according to the business rules corresponding to the first tag group;

[0028] Determine a reward matrix corresponding to each first label according to the determined conflict relationship;

[0029] For each other first tag, determine the reward that the first tag transfers to the other first tag according to the reward matrix;

[0030] Determine a reward relationship value according to the sum of the initial relationship value of the other first tag and the reward corresponding to the other first tag;

[0031] The sub-relationship value with the other first tag is determined by multiplying the reward relationship value of the other first tag by the product of the relationship transfer probability from the first tag to the other first tag.

[0032] Optionally, determining the relationship between the first tags according to the relationship values ​​of the first tags specifically includes:

[0033] For each first tag and each other first tag, in the process of the first tag transmitting the relationship to the other first tag via different second tags, determine the cumulative sum of the relationship values ​​of the other first tags as the value of the tag pair composed of the two first tags;

[0034] The relationship between the two tags is determined according to the sum of the values ​​of the two tags and the value of the tag pair.

[0035] Optionally, determining the relationship between the two tags according to the sum of the values ​​of the two tags and the value of the tag pair specifically includes:

[0036] Weighting the sum of the values ​​of the two tags according to a preset first weight to determine the weighted sum of the values ​​of the two tags;

[0037] Weighting the value of the tag pair according to a preset second weight to determine the weighted value of the tag pair;

[0038] The sum of the weighted value and the weighted value of the tag pair is determined as the relationship between the two tags.

[0039] This specification provides a label relationship determination device based on bipartite graph relationship transfer, including:

[0040] A relationship weight module, used to determine the relationship weight between each first tag in the first tag group and each second tag in the second tag group;

[0041] A one-way transfer probability module, used to determine, according to the relationship weights between each first tag and each second tag, a first probability of relationship transfer from each first tag to each second tag, and a second probability of relationship transfer from each second tag to each first tag;

[0042] A two-way transfer probability module is used to determine, for each first tag, the relationship transfer probabilities of the first tag transferring the relationship to each other first tag via different second tags according to each first probability corresponding to the first tag and each second probability corresponding to each second tag;

[0043] A relationship value module, used to determine the relationship value of the first tag according to the preset initial relationship values ​​of other first tags and the relationship transmission probabilities corresponding to the first tag, wherein the higher the relationship value, the higher the probability that the first tag has an association relationship with other first tags;

[0044] The tag relationship module is used to update the relationship value of each first tag according to the relationship value of the first tag and the corresponding relationship transmission probability until the iteration termination condition is met, and determine the relationship between each first tag according to the relationship value of each first tag.

[0045] This specification provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the above-mentioned label relationship determination method based on bipartite graph relationship transfer is implemented.

[0046] This specification provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the method for determining a label relationship based on bipartite graph relationship transfer is implemented.

[0047] At least one of the above technical solutions adopted in this specification can achieve the following beneficial effects:

[0048] In the label relationship determination method based on bipartite graph relationship transfer provided in this specification, first, according to the relationship weights between each first label in the first label group and each second label in the second label group, the relationship transfer probabilities of the first label transferring the relationship to each other first label through different second labels can be determined for each first label, and then, according to the preset initial relationship values ​​of each other first label and the relationship transfer probabilities corresponding to the first label, the relationship value of the first label is determined, and finally, according to the relationship value of each first label, the relationship between each first label is determined. By determining the relationship transfer situation between each first label, the relationship value of each first label when the relationship transfer is performed is obtained, and the relationship between labels with indirect relationship can be determined, which improves the recall rate of recall based on label relationship. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0050] Figure 1 A schematic diagram of the relationship weights between the tags in the two tag groups provided in this specification;

[0051] Figure 2 A flowchart of a label relationship determination method based on bipartite graph relationship transfer provided in this specification;

[0052] Figure 3 A schematic diagram of a label relationship determination device based on bipartite graph relationship transfer provided in this specification;

[0053] Figure 4 A schematic diagram of an electronic device for implementing a label relationship determination method based on bipartite graph relationship transfer provided in this specification. DETAILED DESCRIPTION

[0054] In order to make the purpose, technical solutions and advantages of this specification more clear, the technical solutions of this application will be clearly and completely described below in combination with the specific embodiments of this specification and the corresponding drawings. Obviously, the described embodiments are only part of the embodiments of this application, not all of them. Based on the embodiments in the specification, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0055] At present, when recalling based on tags, the relationship between tags in the tag group to which the tag belongs is first determined. The tags here can be the server extracting keywords from data from multiple sources through natural language processing algorithms such as keyword recognition, and using the extracted keywords as tags. Alternatively, tags are determined from data from multiple sources based on data mining algorithms, etc. Among them, data from multiple sources include data from user-generated content, professionally produced content, and other sources.

[0056] When determining the relationship between tags in a tag group, first determine the auxiliary tag group corresponding to the tag group. Usually, the auxiliary tag group consists of tags that have an association relationship with the tags in the tag group. Taking the entertainment tag group as an example, people of different age groups, different occupations, and other groups with different attribute tags may have different entertainment activities. It can be seen that there is an association relationship between the attribute tag and the entertainment tag. Therefore, when determining the relationship between tags in the entertainment tag group, the attribute tag group can be used as the auxiliary tag group.

[0057] Then, for each label in the label group, the weight of the label and each label in the auxiliary label group can be determined according to the co-occurrence of the label and each label in the auxiliary label group in the data from multiple sources. The larger the weight, the higher the possibility that the two labels co-occur in the data from multiple sources.

[0058] Finally, the weight vector of the label for the auxiliary label group can be constructed according to the weights of the label and each label in the auxiliary label group, and then the relationship between the two labels can be determined according to the similarity of the weight vectors of any two labels in the label group. When determining the similarity of the weight vector, it can be obtained by calculating the cosine similarity of the two vectors (i.e., the cosine value between the angles of the two vectors). The closer the cosine value is to 1, the closer the angle of the weight vector is to 0, that is, the more similar the two vectors are.

[0059] However, the relationship between the tags in the tag group depends on the co-occurrence of the tags in the tag group and the tags in the auxiliary tag group. Only when two tags co-occur with at least one tag in the auxiliary tag group can the two tags be determined to have a relationship. However, there may be indirect relationships between the tags in the tag group, such as Figure 1 shown.

[0060] Figure 1 This is a diagram showing the weight of the relationship between the labels in the two label groups provided in this specification. Figure 1 It can be seen that tag group A includes tags {a1, a2, a3, a4, a5}, and tag group B includes tags {b1, b2, b3, b4, b5}. Taking tag a3 as an example, the solid line between tag a3 and the tags in tag group B indicates that there is co-occurrence between the two tags, and the relationship weight is not zero. The value on the solid line indicates the relationship weight between the two tags, and the dotted line indicates that there is no co-occurrence between the two tags, and the relationship weight is zero. For the convenience of description, Figure 1 Only the relationship between label a3 and all labels in label group B is drawn for example, and only solid lines when the relationship weights are not zero are drawn between the labels in the other two label groups, and the dotted lines are omitted.

[0061] From the above, we can see that label a1 co-occurs with label b1 and label b3. The weight vector for label group B is [0.8, 0, 0.8, 0, 0], and label a3 co-occurs with label b2. The weight vector for label group B is [0, 0.8, 0, 0, 0], that is, label a1 and label a3 co-occur with different labels in label group B. At this time, it can be calculated that the cosine similarity between label a1 and label a3 is zero, that is, through the cosine similarity between the two label weight vectors, it can only be obtained that label a1 has no relationship with label a3. Therefore, when recalling based on label a1, label a3 will not be recalled.

[0062] However, label a1 co-occurs with label b1, label b1 co-occurs with label a2, label a2 co-occurs with label b2, and label b2 co-occurs with label a3. Therefore, there is an indirect association between label a1 and label a3, and when recalling based on label a1, label a3 should be recalled. However, when determining the relationship between two labels through the cosine similarity between the two label weight vectors, the indirect relationship between the two labels is not calculated, resulting in a low recall rate for recall based on label relationships.

[0063] The technical solutions provided by various embodiments of the present application are described in detail below in conjunction with the accompanying drawings.

[0064] Figure 2 This is a flow chart of a method for determining a label relationship based on bipartite graph relationship transfer in this specification, which specifically includes the following steps:

[0065] S100: Determine the relationship weight between each first tag in the first tag group and each second tag in the second tag group.

[0066] In practical applications, when determining the relationship between tags in a tag group, the auxiliary tag group can be used to determine the association between each tag in the tag group and each tag in the auxiliary tag group, so as to calculate the situation in which the relationship between each tag in the tag group is transmitted through each tag in the auxiliary tag group to construct an indirect relationship, thereby improving the accuracy of determining the relationship between each tag in the tag group.

[0067] In one or more embodiments of the present specification, when determining the relationship between tags in a tag group, the server of the business platform may first determine the relationship weight between each first tag in the first tag group and each second tag in the second tag group. The first tag group refers to the target tag group whose relationship between tags is to be determined, each first tag is each tag in the first tag group, the second tag group refers to the auxiliary tag group, and each second tag is each tag in the auxiliary tag group. For the convenience of description, the following is Figure 1 For example, Figure 1 The tag group A in the figure is the first tag group, and the tag group B is the second tag group. The relationship weight mentioned here may depend on the specific weight algorithm.

[0068] For example, the number of times each first label and each second label co-appear in the multi-source data is counted, and each statistical data can be used as the corresponding relationship weight, or in order to facilitate subsequent calculations, each statistical data can be normalized (such as standard normalization, maximum and minimum normalization), and each normalized data is used as the corresponding relationship weight. For another example, the relationship between each label in the multi-source data is learned through the word vector (word2vector) algorithm, and the weight vector of the label is constructed, and the cosine similarity between the weight vectors between the labels is used as the corresponding relationship weight. Other methods for determining weights are no longer given one by one here, and this specification does not limit how to determine the relationship weights between each first label in the first label group and each second label in the second label group.

[0069] The server mentioned in this specification may refer to a server set up on a business platform, or a device such as a desktop computer or a laptop computer that can execute the solution of this specification. Each label may be a label obtained by the server through a natural language processing algorithm. For the convenience of explanation, the following description is based on the server as the execution subject.

[0070] S102: According to the relationship weights between the first tags and the second tags, respectively determine the first probability of the relationship transfer from the first tags to the second tags, and respectively determine the second probability of the relationship transfer from the second tags to the first tags.

[0071] After determining the relationship weights between each first tag and each second tag as described above, in one or more embodiments of the present specification, the server can determine, based on the relationship weights, a first probability of each first tag transferring a relationship to each second tag, and a second probability of each second tag transferring a relationship to each first tag, so that when calculating the relationship transfer between each tag in the tag group through each tag in the auxiliary tag group, the relationship between the tags in the tag group can be determined more accurately.

[0072] Specifically, for the first probability, the server may first determine the cumulative sum of the relationship weights between the first tag and each second tag for each first tag as the first total weight, and then determine the first probabilities of the first tag transferring the relationship to different second tags according to the ratio of the relationship weights between the first tag and different second tags to the first total weight.

[0073] The first probability can be calculated using the following formula:

[0074]

[0075] In the formula, a i represents the first label, b k represents the second label, ω(a i ,b k ) indicates label a i With label b k The relationship weight, Indicates label a i The cumulative sum of the relationship weights between the second labels in the second label group is the first total weight. i With label b k The ratio of the relationship weight to the first total weight can get the label a i To label b k The first probability of performing relationship transfer, that is, the first probability of performing relationship transfer from the first tag to the second tag.

[0076] For example, for Figure 1 In the first label group, label a1 is in the first label group. Taking label b1 in the second label group as an example, the first probability of label a1 transferring the relationship to label b1 can be expressed as For the calculation method of each second probability of relationship transfer from tag a1 to each other second tag, and each second probability of relationship transfer from each other first tag to each other second tag, please refer to the above corresponding description, and this specification will not repeat it again.

[0077] In one or more embodiments of the present specification, for the second probability, the server may first determine the cumulative sum of the relationship weights between the second tag and each other first tag for each second tag and each first tag as the second total weight, and then determine the second probabilities of the second tag transferring the relationship to each other first tag according to the ratio of the relationship weights between each other first tag and the second tag to the second total weight.

[0078] The second probability can be calculated using the following formula:

[0079]

[0080] In the formula, a i Indicates the first label, a j Indicates that except label a i Other first labels besides b k represents the second label, ω(a j ,b k ) indicates label a j With label b k The relationship weight, Indicates label b k With the first label group except label a i The sum of the relationship weights of all other first labels except a is the second total weight. When calculating the second total weight here, no label a is added. i With label b k The relationship weight. This means that when calculating the label a i The corresponding label b k To label a j When performing the second probability of relationship transfer, label a is not calculated i Label b k To label a i In the case of relationship transfer, this is to comply with the rule that relationship transfer will not be transferred from a certain label to the label itself through other labels. Finally, by calculating the label a j With label b k The ratio of the relationship weight to the second total weight can get the label a i The corresponding label b k To label a j The second probability of performing relationship transfer, that is, the second probability of the second tag performing relationship transfer to the other first tag.

[0081] It should be noted that the label a can be set i Label b k To label a i The second probability of relation transfer is zero.

[0082] For example, for Figure 1 In the first label group, label a1 is in the label group, and label b1 is in the second label group. Taking label a2 in the first label group as an example, the second probability of label a1 transferring the relationship to label a2 via label b1 can be expressed as For the calculation method of each second probability of relationship transfer from tag b1 to each other first tag, and each second probability of relationship transfer from each other second tag to each other first tag, please refer to the above corresponding description, and this specification will not repeat it again.

[0083] This specification does not limit the order of determining the first probability and the second probability, nor does it limit the order of determining the first probabilities corresponding to the first tags, and the order of determining the second probabilities corresponding to the second tags.

[0084] S104: For each first tag, determine, according to each first probability corresponding to the first tag and each second probability corresponding to each second tag, each relationship transfer probability of the first tag transferring the relationship to each other first tag via different second tags.

[0085] After determining the first probabilities corresponding to the first tags and the second probabilities corresponding to the second tags as described above, in one or more embodiments of the specification, the server can determine the relationship transfer probabilities of the first tag to other first tags via different second tags according to the first probabilities corresponding to the first tags and the second probabilities corresponding to the second tags for each first tag. In subsequent steps, the relationship transfer conditions of the first tags to other first tags via different second tags can be determined according to the relationship transfer probabilities, and the relationship value of each first tag can be determined.

[0086] Specifically, the server may determine the sub-relationship transfer probabilities of the first tag corresponding to each second tag based on the product of the corresponding first probability of the first tag transferring the relationship to each second tag and the corresponding second probability of the second tag transferring the relationship to each first tag. According to the sum of the sub-relationship transfer probabilities, the server may determine the relationship transfer probabilities of the first tag transferring the relationship to each other first tag via different second tags.

[0087] The relationship transmission probability can be calculated using the following formula:

[0088]

[0089] In the formula, a i Indicates the first label, a j Indicates that except label a i Other first labels besides b k represents the second label, π(b k |ai ) indicates label a i To label b k The first probability of relation transfer, Indicates label a i The corresponding label b k To label a j The second probability of relationship transfer. Indicates label a i Label b k To label a j The probability of sub-relationship transfer for relationship transfer is the probability of sub-relationship transfer from the first tag to the other first tag via the second tag. Finally, the sub-relationship transfer probabilities corresponding to different second tags in the second tag group are accumulated, and the accumulated sum is used as the label a i To tag a via different second tags j The probability of relationship transfer for relationship transfer.

[0090] For example, for Figure 1 In the first label group, the label a1 and the label a2, the probability of label a1 transferring to label a2 through different second labels can be expressed as For the calculation method of the probability of each relationship transmission from tag a1 to other first tags via different second tags, and the probability of each relationship transmission from other first tags to other first tags via different second tags, please refer to the corresponding description above and this specification will not repeat them again.

[0091] S106: Determine the relationship value of the first tag according to the preset initial relationship values ​​of other first tags and the relationship transmission probabilities corresponding to the first tag, wherein the higher the relationship value, the higher the probability that the first tag has an association relationship with other first tags.

[0092] S108: updating the relationship value of each first tag according to the relationship value of each first tag and the corresponding relationship transmission probability until the iteration termination condition is met, and determining the relationship between each first tag according to the relationship value of each first tag.

[0093] In one or more embodiments of the present specification, after determining, for each first tag, the probability of each relationship transfer from the first tag to other first tags via different second tags, the server can further determine the relationship value of the first tag through iterative calculation, so as to fully calculate the relationship transfer status between the first tags during the iteration process, and in the process of updating the value, the relationship transfer status between the tags will also be transferred through the value.

[0094] During the iterative calculation process, the server can determine the relationship value of the first tag based on the preset initial relationship values ​​of other first tags and the relationship transmission probabilities corresponding to the first tag. And update the relationship value of each first tag based on the relationship value of the first tag and the corresponding relationship transmission probabilities until the iteration termination condition is met. Of course, the server can determine whether the iteration termination condition is met after executing each round of iterative calculation. If the iteration termination condition is not met, the server will continue to execute the steps of the subsequent iterative process. If it is met, the server will stop the calculation and determine the relationship value of each first tag obtained in this round of iterative calculation.

[0095] Among them, the preset initial relationship value can be an arbitrary constant. The relationship value mentioned here represents the possibility that each first tag has an association relationship with each other first tag. The higher the relationship value, the higher the probability that the corresponding first tag has an association relationship with each other first tag.

[0096] The above-mentioned iteration termination condition can be set according to the actual situation. For example, the number of iteration rounds (such as 10 rounds) can be preset according to the business situation corresponding to the first tag group. When the number of iteration rounds is greater than the number of iteration rounds, the iteration is terminated. A minimum update threshold can also be preset according to the above-mentioned relationship weight. If the average value of the difference between the initial relationship value and the updated relationship value of each first tag in the current round is less than the minimum update threshold, the iteration is terminated, and so on. Other ways of setting the iteration termination condition are not given one by one in this manual. Multiple iteration termination conditions can also be used in combination as needed. In this manual, there is no limitation on the specific iteration termination condition.

[0097] Specifically, during the iterative calculation process, the server can determine the sub-relationship value for each other first tag according to the product of the preset initial relationship value of the other first tag and the relationship transfer probability of the first tag transferring the relationship to the other first tag, and determine the relationship value of the first tag according to the sum of the sub-relationship values ​​of the first tag.

[0098] The relationship value of the first label can be calculated using the following formula:

[0099]

[0100] In the formula, a i Indicates the first label, a j Indicates that except label a i Other first labels besides b k Represents the second label, V(a i ) indicates label a i The relationship value to be updated in the current round, V(a j ) indicates label a j The initial relationship value of the current round, Indicates label a i Transfer relationship to tag a via different second tags j The probability of relationship transmission is That is label a i Transfer relationship to tag a via different second tags j The sub-relationship value of the first label group except label a i The sub-relationship values ​​corresponding to the first label other than are accumulated, and the accumulated sum is used as the label a of the current round i relationship value.

[0101] For example, for Figure 1 For label a1 in the first label group in , for each iteration round, the relation value of label a1 can be expressed as For the calculation method of the relationship value corresponding to each other first tag, please refer to the corresponding description above, and this specification will not go into details.

[0102] Then, in one or more embodiments of the present specification, after the server obtains the relationship values ​​of the first tags as described above, because when determining the relationship values, the server has already calculated the situation where there is an indirect relationship between the first tags through relationship transmission, so the relationship between the first tags can be determined according to the relationship values ​​of the first tags.

[0103] Specifically, for each first tag and each other first tag, in the process of the first tag transmitting the relationship to the other first tag via different second tags, the server can determine the cumulative sum of the relationship values ​​of the other first tags as the value of the tag pair formed by the two first tags. And determine the relationship between the two tags according to the sum of the values ​​of the two tags and the value of the tag pair.

[0104] The relationship between two labels can be calculated using the following formula:

[0105]

[0106] stb k ∈B,ω(a i ,b k )≠0,ω(a j ,b k )≠0

[0107] In the formula, a i Indicates the first label, a j Indicates that except label a i Other first labels besides b k Represents the second label, V(a i ) indicates label a i The relationship value, V(a j) indicates label a j The relationship value, ∑V(a j ) indicates label a i With label a j The value that makes up the tag pair. stb k ∈B,ω(a i ,b k )≠0,ω(a j ,b k )≠0 represents the constraint condition when calculating the relationship between two labels, that is, for each label b k , if label a i With label b k There is a co-occurrence and label b k With label a j There is a co-occurrence situation, that is, label a i Can be labeled k Pass the relation to label a j , then for label a j The relationship value is accumulated once, and the accumulated sum is used as label a i With label a j The value of the label pair is determined by label a i The relationship value, label a j The sum of the relationship value and the value of the label pair is taken as label a i With label a j The relationship between.

[0108] For example, for Figure 1 The relationship between the label a1 and the label a2 in the first label group can be expressed as:

[0109] S(a1,a2)=V(a1)+V(a2)+∑V(a2)

[0110] stb k ∈B,ω(a1,b k )≠0,ω(a2,b k )≠0

[0111] For the calculation method of the relationship between the other two first tags, please refer to the corresponding description above, and this specification will not go into details.

[0112] The above method first determines the relationship value of each first tag by calculating the relationship transmission between each first tag, and finally determines the relationship between each first tag according to the relationship value of each first tag, thereby ensuring the accuracy of determining the relationship between tags.

[0113] based on Figure 2The method for determining label relationships based on bipartite graph relationship transfer shown in the figure first determines, for each first label, the relationship weights between each first label in the first label group and each second label in the second label group, and the first probabilities corresponding to the first label are determined according to the relationship weights between the first label and different second labels, and the ratio of the total weights of the first label and each second label, and the probability of the first label transferring the relationship to different second labels is represented by the first probability. For each first label and each first label, the relationship weights between each other first label and the second label are determined, and the ratio of the total weights of the second label and each other first label is determined, and the second probabilities corresponding to the second label are determined, and the probability of the second label corresponding to the first label transferring the relationship to each other first label is represented by the second probability. Then, the relationship transfer probabilities of each first label can be determined according to each first probability and each second probability. The relationship transfer probability represents the probability that the first tag transfers the relationship to each other first tag through different second tags. Then, the relationship value of the first tag is determined according to the preset initial relationship values ​​of each other first tag and the relationship transfer probabilities corresponding to the first tag. Finally, the relationship between each first tag is determined according to the relationship values ​​of each first tag.

[0114] By calculating the situations in which the relationship between each first label is transferred via different second labels, the relationship value of each first label when the relationship is transferred is obtained. This allows the relationship between labels with indirect relationship to be determined based on the relationship transfer probability between each first label via different second labels, even if there is no co-occurrence between the two first labels and any second label in the second label group. This improves the recall rate based on label relationship.

[0115] Based on this, by applying the label relationship determination method based on bipartite graph relationship transmission provided in this specification, the server can calculate the indirect relationship between two labels in the label group, determine a more accurate relationship between labels, and improve the recall rate when recalling based on label relationships. Specific services can be applied on this basis. For example, an entertainment service APP can recall more related labels for a certain label of a certain entertainment project based on the relationship between labels determined by the server, so as to more accurately recommend entertainment projects to users.

[0116] In addition, in one or more embodiments of the present specification, in step S106 and step S108, when the server determines the relationship value of each first tag, in order to reflect the process of the first tag transferring the relationship to other first tags through different second tags, the server can first weight the initial relationship value of each first tag according to the preset attenuation weight to determine the attenuation relationship value of each first tag, and then, for each other first tag, determine the sub-relationship value according to the attenuation relationship value of each first tag and the product of the relationship transfer probability of the first tag transferring the relationship to the other first tag. The attenuation weight is a value not less than zero and not greater than one, which can be set according to actual business needs. The specific value of the attenuation weight is not limited in the present specification. The attenuation relationship value here reflects the indirectness of the association relationship in the relationship transfer process.

[0117] The sub-relationship value can be calculated using the following formula:

[0118]

[0119] In the formula, a i Indicates the first label, a j Indicates that except label a i Other first labels besides b k represents the second label, γ represents the decay weight, γV(a j ) indicates label a j The decay relationship value of the current round, the meaning of the other various formulas can refer to the corresponding descriptions in step S106 and step S108, and this manual will not repeat them. By reflecting the indirectness of the association relationship in the relationship transfer, the initial relationship value of each first tag is weighted according to the decay weight, so that a more accurate sub-relationship value can be determined.

[0120] For example, for Figure 1 In the first tag group of label a1, for each iteration round, the value of the sub-relationship that label a1 transfers to label a2 can be expressed as:

[0121]

[0122] Furthermore, in one or more embodiments of the present specification, in step S106 and step S108, when the server determines the relationship value of each first tag, it can also reflect the special relationship between the businesses. For example, businesses that are in conflict with each other or businesses that are strongly associated with each other can determine the sub-relationship values ​​of the first tags based on the pre-set relationships between the first tags.

[0123] Specifically, first, the first tags that are in conflict with each other can be determined according to the business rules corresponding to the first tag group, and the reward matrix corresponding to each first tag can be determined according to the determined conflicting relationship. Then, for each other first tag, the reward for the first tag transferring the relationship to the other first tag is determined according to the reward matrix, and the reward relationship value is determined according to the sum of the initial relationship value of the other first tag and the reward corresponding to the other first tag. Finally, the sub-relationship value with the other first tag is determined according to the reward relationship value of the other first tag multiplied by the product of the relationship transfer probability of the first tag transferring the relationship to the other first tag. Among them, in the reward matrix, the value of the reward for each first tag transferring the relationship to each other first tag can be specifically set according to the actual business situation, and the specific value of the reward is not limited in this specification.

[0124] For example, if the relationship weights between each first label and each second label are normalized and distributed between zero and one, it can be determined that the reward for relationship transfer between first labels that have no special relationship is zero, and the reward for relationship transfer between first labels that have a conflicting relationship is negative one. Of course, for first labels that have a highly correlated relationship with each other, the reward for relationship transfer can also be set to positive one. Through the reward matrix, rewards can be further set for relationship transfer between first labels with special relationships.

[0125] As shown in Table 1, a1, a2, and a3 represent three different first labels. In the table, the first row and the first column correspond to the first label, and the reward matrix is ​​represented except for the first row and the first column. There is no special relationship between label a1 and label a2, and the reward for relationship transfer is 0. There is a high correlation relationship between label a1 and label a3, and the reward for relationship transfer is 1. There is a conflicting relationship between label a2 and label a3, and the reward for relationship transfer is -1. It should be noted that relationship transfer will not transfer the relationship from a certain label to the label itself, and the reward for relationship transfer between the label and itself can be set to 0.

[0126]

[0127] Table 1 can use the following formula to calculate the value of the sub-relationship:

[0128]

[0129] In the formula, a i Indicates the first label, a j Indicates that except label a i Other first labels besides b k Indicates the second label, Indicates label a iTransfer relation to label a j The meanings of the other types of rewards can be found in the corresponding descriptions in step S106 and step S108, and will not be repeated in this manual.

[0130] for Figure 1 In the first tag group of label a1, for each iteration round, the value of the sub-relationship that label a1 transfers to label a2 can be expressed as:

[0131]

[0132] Furthermore, in one or more embodiments of the present specification, in step S106 and step S108, when the server determines the relationship value of each first tag, it can also simultaneously reflect the process of the first tag transmitting the relationship to other first tags through different second tags, the indirectness of the association relationship, and the special relationship between the businesses.

[0133] First, the initial relationship value of each first tag is weighted according to the preset decay weight to determine the decay relationship value of each first tag. Then, for each other first tag, the reward for the first tag to transfer the relationship to the other first tag is determined according to the preset reward matrix, and the reward relationship value is determined according to the decay relationship value of the other first tag and the sum of the reward for the first tag to transfer the relationship to the other first tag. Finally, the sub-relationship value with the other first tag is determined according to the reward relationship value of the other first tag multiplied by the relationship transfer probability of the first tag to transfer the relationship to the other first tag. This reflects the indirectness of the association relationship in the relationship transfer process and realizes the needs of business customization.

[0134] The sub-relationship value can be calculated using the following formula:

[0135]

[0136] In the formula, a i Indicates the first label, a j Indicates that except label a i Other first labels besides b k represents the second label, γ represents the decay weight, γV(a j ) indicates label a j The decay relationship value of the current round, Indicates label a i Transfer relation to label a j The meanings of the other types of rewards can be found in the corresponding descriptions in step S106 and step S108, and will not be repeated in this manual.

[0137] For example, for Figure 1In the first tag group of label a1, for each iteration round, the value of the sub-relationship that label a1 transfers to label a2 can be expressed as:

[0138]

[0139] In addition, in one or more embodiments of the present specification, when the server determines the value of the label pair consisting of the two first labels for each first label and each other first label, it can also determine the value of the label pair consisting of the two first labels according to the cumulative sum of the attenuated relationship values ​​of the other first labels in the process of the first label transferring the relationship to the other first labels via different second labels, so as to reflect the indirectness of the association relationship in the relationship transfer process.

[0140] The value of a label pair can be calculated using the following formula:

[0141] Q(a i ,a j )=γ∑V(a j )

[0142] stb k ∈B,ω(a i ,b k )≠0,ω(a j ,b k )≠0

[0143] In the formula, a i Indicates the first label, a j Indicates that except label a i Other first labels besides b k represents the second label, γ represents the attenuation weight, and the meanings of the remaining formulas can be referred to the corresponding descriptions in step S106 and step S108, which will not be repeated in this specification.

[0144] For example, for Figure 1 The value of the label pair consisting of label a1 and label a2 in the first label group in can be expressed as:

[0145] Q(a1,a2)=γ∑V(a2)

[0146] stb k ∈B,ω(a1,b k )≠0,ω(a2,b k )≠0

[0147] Furthermore, in one or more embodiments of the present specification, when the server determines the value of the tag pair consisting of the two first tags for each first tag and each other first tag, it can also determine the value of the tag pair consisting of the two first tags based on the cumulative sum of the relationship values ​​of the other first tags and the sum of the rewards for transferring the relationship from the first tag to the other first tags during the process of the first tag transferring the relationship to the other first tags via different second tags, so as to realize the needs of business customization, that is, to reduce the relationship between first tags that are in conflict with each other, and of course, to enhance the relationship between first tags that are highly correlated with each other.

[0148] The value of a label pair can be calculated using the following formula:

[0149]

[0150] stb k ∈B,ω(a i ,b k )≠0,ω(a j ,b k )≠0

[0151] In the formula, a i Indicates the first label, a j Indicates that except label a i Other first labels besides b k Indicates the second label, Indicates label a i Transfer relation to label a j The meanings of the other types of rewards can be found in the corresponding descriptions in step S106 and step S108, and will not be repeated in this manual.

[0152] For example, for Figure 1 The value of the label pair consisting of label a1 and label a2 in the first label group in can be expressed as:

[0153]

[0154] stb k ∈B,ω(a1,b k )≠0,ω(a2,b k )≠0

[0155] Furthermore, in one or more embodiments of the present specification, when the server determines the value of the tag pair consisting of the two first tags for each first tag and each other first tag, it can also determine the value of the tag pair consisting of the two first tags according to the cumulative sum of the attenuation relationship values ​​of the other first tags and the sum of the rewards of the first tag transferring the relationship to the other first tags in the process of the first tag transferring the relationship to the other first tags via different second tags, so as to reflect the indirectness of the association relationship in the relationship transfer process and realize the needs of business customization, that is, to reduce the relationship between first tags that are in conflict with each other, and of course, to enhance the relationship between first tags that are highly associated with each other.

[0156] The value of a label pair can be calculated using the following formula:

[0157]

[0158] stb k ∈B,ω(a i ,b k )≠0,ω(a j ,b k )≠0

[0159] In the formula, a i Indicates the first label, a j Indicates that except label a i Other first labels besides b k represents the second label, γ represents the decay weight, Indicates label a i Transfer relation to label a j The meanings of the other types of rewards can be found in the corresponding descriptions in step S106 and step S108, and will not be repeated in this manual.

[0160] For example, for Figure 1 The value of the label pair consisting of label a1 and label a2 in the first label group in can be expressed as:

[0161]

[0162] stb k ∈B,ω(a1,b k )≠0,ω(a2,b k )≠0

[0163] In addition, in one or more embodiments of the present specification, when determining the relationship between the two tags based on the value of the two tags and the sum of the values ​​of the tag pair formed by the two tags, the value of the two tags and the value of the tag pair may not be in the same order of magnitude. In order to make the value of the two tags and the value of the tag pair be calculated in the same order of magnitude, the sum of the values ​​of the two tags may be weighted according to a preset first weight to determine the weighted value sum of the two tags, and then the value of the tag pair may be weighted according to a preset second weight to determine the weighted value of the tag pair, and finally, the sum of the weighted value and the weighted value of the tag pair may be determined as the relationship between the two tags. The preset first weight and the preset second weight may be set according to the actual business situation, and the specific values ​​of the two weights are not limited in the present specification.

[0164] Assume label a i With label a j The value of the label pair is ∑V(a j ), the relationship between two labels can be calculated using the following formula:

[0165] S(a i ,a j )=α1{V(a i )+V(a j )}+α2∑V(a j )

[0166] stb k ∈B,ω(a i ,b k )≠0,ω(a j ,b k )≠0

[0167] In the formula, a i Indicates the first label, a j Indicates that except label a i Other first labels besides b k represents the second label, α1 represents the preset first weight, α2 represents the preset second weight, and the meanings of the remaining formulas can be referred to the corresponding descriptions in step S106 and step S108, which will not be repeated in this specification.

[0168] Among them, for label a i With label a j When the value of the composed label pair is any of the above-mentioned cases, the expression form of the value of the label pair can be adjusted accordingly, and examples will not be given one by one here.

[0169] For example, for Figure 1In the first tag group, the tag a1 and the tag a2 are in the same group. When the value of the tag pair consisting of the tag a1 and the tag a2 is ∑V(a2), the relationship between the tag a1 and the tag a2 can be expressed as:

[0170] S(a1,a2)=α1{V(a1)+V(a2)}+α2∑V(a2)

[0171] stb k ∈B,ω(a1,b k )≠0,ω(a2,b k )≠0

[0172] In addition, in one or more embodiments of the present specification, when determining the relationship between the two tags based on the value of the two tags and the sum of the values ​​of the tag pair consisting of the two tags, in order to facilitate practical application, the relationship between the two tags can also be adjusted so that the relationship between the first tag and other first tags is equal to the relationship between the other first tags and the first tag.

[0173] Specifically, the relationship between the two labels can be determined according to the average value of the relationship between the two labels, so that the relationship between the labels satisfies symmetry. The relationship between the two labels can be adjusted using the following formula:

[0174]

[0175] In the formula, a i Indicates the first label, a j Indicates that except label a i Other first labels besides b k represents the second label, S(a i ,a j ) represents the adjusted label a i With label a j The relationship between S(a j ,a i ) represents the adjusted label a j With label a i The relationship between S′(a i ,a j ) indicates that according to label a i The value of label a j The value and label a i With label a j The value of the label pair, the calculated label a i With label a j The relationship between S′(a j ,a i ) indicates that according to label a i The value of label aj The value and label a j With label a i The value of the label pair, the calculated label a j With label a i The relationship between.

[0176] For example, for Figure 1 The labels a1 and a2 in the first label group, the relationship between labels a1 and a2, and the relationship between labels a2 and a1 can be expressed as:

[0177]

[0178] The above is a method provided by one or more embodiments of this specification. Based on the same idea, this specification also provides a corresponding label relationship determination device based on bipartite graph relationship transfer, such as Figure 3 shown.

[0179] Figure 3 A schematic diagram of a label relationship determination device based on bipartite graph relationship transfer provided in this specification includes:

[0180] A relationship weight module 300, for determining a relationship weight between each first tag in the first tag group and each second tag in the second tag group;

[0181] A one-way transfer probability module 302 is used to determine, according to the relationship weights between each first tag and each second tag, a first probability of relationship transfer from each first tag to each second tag, and a second probability of relationship transfer from each second tag to each first tag;

[0182] The bidirectional transfer probability module 304 is used to determine, for each first tag, the relationship transfer probabilities of the first tag transferring the relationship to each other first tag via different second tags according to each first probability corresponding to the first tag and each second probability corresponding to each second tag;

[0183] The relationship value module 306 is used to determine the relationship value of the first tag according to the preset initial relationship values ​​of other first tags and the relationship transfer probabilities corresponding to the first tag, wherein the higher the relationship value, the higher the probability that the first tag has an association relationship with other first tags;

[0184] The tag relationship module 308 is used to update the relationship value of each first tag according to the relationship value of the first tag and the corresponding relationship transmission probability until the iteration termination condition is met, and determine the relationship between each first tag according to the relationship value of each first tag.

[0185] Optionally, the one-way transmission probability module 302 determines, for each first tag, the cumulative sum of the relationship weights between the first tag and each second tag as the first total weight, and determines the first probabilities of the first tag transferring the relationship to different second tags based on the ratio of the relationship weights between the first tag and different second tags to the first total weight.

[0186] Optionally, the one-way transmission probability module 302 determines, for each second tag and each first tag, the cumulative sum of the relationship weights between the second tag and each other first tag as the second total weight, and determines the second probabilities of the second tag transmitting the relationship to each other first tag based on the ratio of the relationship weights between each other first tag and the second tag to the second total weight.

[0187] Optionally, the bidirectional transfer probability module 304 determines the sub-relationship transfer probabilities of the first tag corresponding to each second tag based on the product of the corresponding first probability of relationship transfer from the first tag to each second tag and the corresponding second probability of relationship transfer from each second tag to each first tag, and determines the relationship transfer probabilities of the first tag transferring the relationship to each other first tag via different second tags based on the sum of the sub-relationship transfer probabilities.

[0188] Optionally, the relationship value module 306 determines the sub-relationship value for each other first tag based on a preset initial relationship value of the other first tag and the product of the relationship transfer probability of the first tag transferring the relationship to the other first tag, and determines the relationship value of the first tag based on the sum of the sub-relationship values ​​of the first tag.

[0189] Optionally, the relationship value module 306 determines first tags that are in conflict with each other according to the business rules corresponding to the first tag group, determines a reward matrix corresponding to each first tag according to the determined conflicting relationship, determines for each other first tag according to the reward matrix, determines the reward for the first tag transferring the relationship to the other first tag, determines the reward relationship value according to the sum of the initial relationship value of the other first tag and the reward corresponding to the other first tag, determines the sub-relationship value with the other first tag according to the reward relationship value of the other first tag multiplied by the product of the relationship transfer probability of the first tag transferring the relationship to the other first tag.

[0190] Optionally, the label relationship module 308, for each first label and each other first label, determines the cumulative sum of the relationship values ​​of the other first labels as the value of the label pair consisting of the two first labels in the process of the first label transferring the relationship to the other first label via different second labels, and determines the relationship between the two labels based on the sum of the values ​​of the two labels and the value of the label pair.

[0191] Optionally, the label relationship module 308 weights the sum of the values ​​of the two labels according to a preset first weight to determine the weighted sum of the values ​​of the two labels, weights the values ​​of the label pair according to a preset second weight to determine the weighted value of the label pair, and determines the sum of the weighted values ​​and the sum of the weighted values ​​of the label pair as the relationship between the two labels.

[0192] This specification also provides a computer-readable storage medium, which stores a computer program, which can be used to execute the above Figure 2 A label relationship determination method based on bipartite graph relationship transfer is provided.

[0193] This manual also provides Figure 4 The structural diagram of the electronic device shown in FIG. Figure 4 As mentioned above, at the hardware level, the electronic device includes a processor, an internal bus, a network interface, a memory, and a non-volatile memory, and may also include other hardware required for the business. The processor reads the corresponding computer program from the non-volatile memory into the memory and then runs it to achieve the above Figure 2 The label relationship determination method based on bipartite graph relationship transfer.

[0194] Of course, in addition to software implementation, this specification does not exclude other implementation methods, such as the combination of logic device XOR software and hardware, etc., that is to say, the executor of the following processing flow is not limited to each logic unit, but can also be hardware or logic device.

[0195] In the 1990s, improvements to a technology could be clearly distinguished as hardware improvements (for example, improvements to the circuit structure of diodes, transistors, switches, etc.) or software improvements (improvements to the method flow). However, with the development of technology, many improvements to the method flow today can be regarded as direct improvements to the hardware circuit structure. Designers almost always obtain the corresponding hardware circuit structure by programming the improved method flow into the hardware circuit. Therefore, it cannot be said that an improvement in a method flow cannot be implemented using a hardware entity module. For example, a programmable logic device (PLD) (such as a field programmable gate array (FPGA)) is such an integrated circuit whose logical function is determined by the user's programming of the device. Designers can "integrate" a digital system on a PLD by programming it themselves, without having to ask a chip manufacturer to design and produce a dedicated integrated circuit chip. Moreover, nowadays, instead of manually making integrated circuit chips, this kind of programming is mostly implemented by "logic compiler" software, which is similar to the software compiler used when developing and writing programs, and the original code before compilation must also be written in a specific programming language, which is called hardware description language (HDL). There is not only one HDL, but many kinds, such as ABEL (Advanced Boolean Expression Language), AHDL (Altera Hardware Description Language), Confluence, CUPL (Cornell University Programming Language), HDCal, JHDL (Java Hardware Description Language), Lava, Lola, MyHDL, PALASM, RHDL (Ruby Hardware Description Language), etc. The most commonly used ones are VHDL (Very-High-Speed ​​Integrated Circuit Hardware Description Language) and Verilog. Those skilled in the art should also know that it is only necessary to program the method flow slightly in the above-mentioned hardware description languages ​​and program it into the integrated circuit, and then it is easy to obtain the hardware circuit that implements the logic method flow.

[0196] The controller can be implemented in any appropriate manner, for example, the controller can take the form of a microprocessor or processor and a computer-readable medium storing a computer-readable program code (such as software or firmware) that can be executed by the (micro)processor, a logic gate, a switch, an application-specific integrated circuit (ASIC), a programmable logic controller, and an embedded microcontroller. Examples of controllers include, but are not limited to, the following microcontrollers: ARC 625D, Atmel AT91SAM, Microchip PIC18F26K20, and Silicone Labs C8051F320. The memory controller can also be implemented as part of the control logic of the memory. Those skilled in the art also know that in addition to implementing the controller in a purely computer-readable program code manner, the controller can be implemented in the form of a logic gate, a switch, an application-specific integrated circuit, a programmable logic controller, and an embedded microcontroller by logically programming the method steps. Therefore, this controller can be considered as a hardware component, and the devices included therein for implementing various functions can also be regarded as structures within the hardware component. Or even, the devices for implementing various functions can be regarded as both software modules for implementing the method and structures within the hardware component.

[0197] The systems, devices, modules or units described in the above embodiments may be implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer. Specifically, the computer may be, for example, a personal computer, a laptop computer, a cellular phone, a camera phone, a smart phone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or a combination of any of these devices.

[0198] For the convenience of description, the above device is described in various units according to their functions. Of course, when implementing this specification, the functions of each unit can be implemented in the same or multiple software and / or hardware.

[0199] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0200] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0201] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0202] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0203] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0204] The memory may include non-permanent storage in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. The memory is an example of a computer-readable medium.

[0205] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0206] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0207] Those skilled in the art will appreciate that the embodiments of this specification may be provided as methods, systems or computer program products. Therefore, this specification may take the form of a complete hardware embodiment, a complete software embodiment or an embodiment combining software and hardware. Moreover, this specification may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0208] This specification may be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. This specification may also be practiced in distributed computing environments where tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules may be located in local and remote computer storage media, including storage devices.

[0209] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0210] The above description is only an embodiment of the present specification and is not intended to limit the present specification. For those skilled in the art, the present specification may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present specification shall be included in the scope of the claims of the present specification.

Claims

1. A label relationship determination method based on bipartite graph relationship transfer, characterized in that: include: Determine the relationship weight between each first tag in the first tag group and each second tag in the second tag group; According to the relationship weights between the first tags and the second tags, respectively determine the first probability of the relationship transfer from the first tags to the second tags, and respectively determine the second probability of the relationship transfer from the second tags to the first tags; For each first tag, determine, according to each first probability corresponding to the first tag and each second probability corresponding to each second tag, each relationship transfer probability of the first tag transferring the relationship to each other first tag via different second tags; Determine the relationship value of the first tag according to the preset initial relationship values ​​of other first tags and the relationship transfer probabilities corresponding to the first tag, wherein the higher the relationship value, the higher the probability that the first tag has an association relationship with other first tags; According to the relationship values ​​of the first tags and the corresponding transmission probabilities of each relationship, the relationship values ​​of each first tag are updated until the iteration termination condition is met, and the relationship between each first tag is determined according to the relationship values ​​of each first tag; Determining first probabilities of relationship transfer from each first tag to each second tag respectively includes: For each first tag, determine the cumulative sum of the relationship weights between the first tag and each second tag as the first total weight; Determine first probabilities of transferring the relationship from the first tag to different second tags, respectively, according to the ratio of the relationship weights between the first tag and different second tags to the first total weight; Determining the second probability of each second tag transferring the relationship to each first tag respectively includes: For each second tag and each first tag, determine the cumulative sum of the relationship weights between the second tag and each other first tag as the second total weight; Second probabilities of the second tag transferring the relationship to other first tags are determined respectively according to the ratio of the relationship weights between other first tags and the second tag to the second total weight.

2. The method according to claim 1, characterized in that Determining the probability of transferring the relationship of the first tag to other first tags via different second tags specifically includes: Determine each sub-relationship transfer probability of the first tag corresponding to each second tag according to the product of the corresponding first probability of the first tag transferring the relationship to each second tag and the corresponding second probability of the second tag transferring the relationship to each first tag; According to the sum of the sub-relationship transfer probabilities, the relationship transfer probabilities of the first tag transferring the relationship to other first tags via different second tags are determined.

3. The method according to claim 1, characterized in that Determining the relationship value of the first tag according to the preset initial relationship values ​​of other first tags and the relationship transfer probabilities corresponding to the first tag specifically includes: For each other first tag, determine the sub-relationship value according to the product of the preset initial relationship value of the other first tag and the relationship transfer probability of the first tag transferring the relationship to the other first tag; The relationship value of the first tag is determined according to the sum of the sub-relationship values ​​of the first tag.

4. The method according to claim 3, characterized in that Determining the sub-relationship value according to the preset initial relationship value of the other first tag and the product of the relationship transfer probability of the first tag transferring the relationship to the other first tag specifically includes: Determining first tags that conflict with each other according to the business rules corresponding to the first tag group; Determine a reward matrix corresponding to each first label according to the determined conflict relationship; For each other first tag, determine the reward that the first tag transfers to the other first tag according to the reward matrix; Determine a reward relationship value according to the sum of the initial relationship value of the other first tag and the reward corresponding to the other first tag; The sub-relationship value with the other first tag is determined by multiplying the reward relationship value of the other first tag by the product of the relationship transfer probability from the first tag to the other first tag.

5. The method according to claim 1, characterized in that According to the relationship values ​​of the first tags, the relationship between the first tags is determined, specifically including: For each first tag and each other first tag, in the process of the first tag transmitting the relationship to the other first tag via different second tags, determine the cumulative sum of the relationship values ​​of the other first tags as the value of the tag pair composed of the two first tags; The relationship between the two tags is determined according to the sum of the values ​​of the two tags and the value of the tag pair.

6. The method according to claim 5, characterized in that Determining the relationship between the two tags according to the sum of the values ​​of the two tags and the value of the tag pair specifically includes: Weighting the sum of the values ​​of the two tags according to a preset first weight to determine the weighted sum of the values ​​of the two tags; Weighting the value of the tag pair according to a preset second weight to determine the weighted value of the tag pair; The sum of the weighted value and the weighted value of the tag pair is determined as the relationship between the two tags.

7. A label relationship determination device based on bipartite graph relationship transfer, characterized in that: include: A relationship weight module, used to determine the relationship weight between each first tag in the first tag group and each second tag in the second tag group; A one-way transfer probability module, used to determine, according to the relationship weights between each first tag and each second tag, a first probability of relationship transfer from each first tag to each second tag, and a second probability of relationship transfer from each second tag to each first tag; Determining the first probability of each first tag transferring the relationship to each second tag respectively, specifically includes: for each first tag, determining the cumulative sum of the relationship weights between the first tag and each second tag as the first total weight, and determining the first probabilities of the first tag transferring the relationship to different second tags according to the ratio of the relationship weights between the first tag and different second tags to the first total weight; determining the second probability of each second tag transferring the relationship to each first tag respectively, specifically includes: for each second tag and each first tag, determining the cumulative sum of the relationship weights between the second tag and each other first tag as the second total weight, and determining the second probabilities of the second tag transferring the relationship to each other first tag according to the ratio of the relationship weights between each other first tag and the second tag to the second total weight; A two-way transfer probability module is used to determine, for each first tag, the relationship transfer probabilities of the first tag transferring the relationship to each other first tag via different second tags according to each first probability corresponding to the first tag and each second probability corresponding to each second tag; A relationship value module, used to determine the relationship value of the first tag according to the preset initial relationship values ​​of other first tags and the relationship transmission probabilities corresponding to the first tag, wherein the higher the relationship value, the higher the probability that the first tag has an association relationship with other first tags; The tag relationship module is used to update the relationship value of each first tag according to the relationship value of the first tag and the corresponding relationship transmission probability until the iteration termination condition is met, and determine the relationship between each first tag according to the relationship value of each first tag.

8. A computer-readable storage medium, characterized in that: The storage medium stores a computer program, and when the computer program is executed by a processor, the method according to any one of claims 1 to 6 is implemented.

9. An electronic device, characterized in that: The method comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the method described in any one of claims 1 to 6 is implemented.

Images