A communication object prediction method and system based on multi-dimensional knowledge attributes

Abstract

The present invention discloses a communication object prediction method and system based on multidimensional knowledge attributes, belonging to the technical field of communication network user behavior analysis. The main purpose is to solve the problems of single modeling of user communication process and insufficient information utilization. The main solution is based on the user's historical communication records containing multidimensional knowledge attributes. First, according to the historical communication records and their multidimensional knowledge attributes, the corresponding high-dimensional space representation and communication object importance sequence are generated; secondly, starting from different levels, the corresponding high-dimensional space representation is spliced to obtain the feature sequence of the corresponding level, and then the attention mechanism is used to extract the corresponding interest from the feature sequence; then, an interest combination mechanism based on multidimensional knowledge attributes is introduced to combine the interest of different levels with the importance of communication objects to form a global feature representation of the communication target; finally, combined with the local feature representation of the communication object, the prediction of the user's call object is realized.

Description

Technical Field

[0001] The present invention relates to the technical field of communication network user behavior analysis, and in particular to a communication object prediction method and system based on multi-dimensional knowledge attributes. Background Art

[0002] Mobile communication tools, especially smartphones, have become an integral part of people's lives, greatly facilitating their daily lives and production. This data, along with mobile communication tools, forms the communication network of today's society. However, these convenient communication tools have also become a means of communication and organization for criminal gangs. To assist law enforcement agencies in combating these criminal gangs, predicting the communication partners of users in these networks has become a crucial issue.

[0003] However, there is not much work currently focusing on this issue. At the same time, most existing work only uses the communication partner's phone number as historical communication data. Although this approach has proven its effectiveness in other transferable fields, it models the user's communication process in a single way, affecting the accuracy of its results. On the other hand, this approach ignores objective factors such as the community to which the communication partner belongs, the duration of the communication, and the importance of the communication partner in the communication network, resulting in insufficient utilization of the user's historical communication record information. To this end, this patent proposes a communication partner prediction method and system based on multidimensional knowledge attributes by integrating the multidimensional knowledge attributes of the communication network, solving the problems of single modeling of the user's communication process and insufficient information utilization. Summary of the Invention

[0004] The purpose of the present invention is to solve the problem that most existing works only use the number of the communication object as historical communication data. Although this approach has been proven to be effective in other transferable fields, it only models the user communication process in a single way, affecting the accuracy of its results. On the other hand, this approach ignores objective factors such as the community to which the communication object belongs, the duration of the communication, and the importance of the communication object in the communication network, resulting in insufficient utilization of user historical communication record information.

[0005] In order to achieve the above-mentioned purpose, the present invention provides a communication object prediction method and system based on multi-dimensional knowledge attributes, which fully utilizes the multi-dimensional knowledge attribute information in the user's historical communication sequence, models the user communication process from the communication object level, communication community level and communication importance level, and realizes the prediction of the communication user's next communication object, thereby solving the problems mentioned in the above-mentioned background technology.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a communication target prediction method and system based on multi-dimensional knowledge attributes, wherein the communication target prediction method comprises the following specific steps:

[0007] Step S10: Based on the user's communication object number, communication object community, communication service type, communication caller relationship, communication object global criticality and communication duration in the user's historical communication records, a corresponding communication object sequence, communication community sequence, communication information operation sequence and communication object importance sequence are constructed, and a corresponding dictionary is constructed;

[0008] Step S20: Initialize all dictionaries separately, form high-dimensional spatial representations of corresponding sequences, use three gated recurrent unit neural networks to model the communication object number, communication object community, and communication information operation, and update the high-dimensional spatial representations of the communication object sequence, communication community sequence, and communication information operation sequence;

[0009] Step S30: Concatenate the high-dimensional spatial representation of the communication object sequence with the high-dimensional spatial representation of the communication information operation sequence to obtain the communication object-information operation sequence, and use the attention mechanism to extract the communication object-level interest from it; use a similar method to concatenate the high-dimensional spatial representation of the communication community sequence with the high-dimensional spatial representation of the communication information operation sequence, and then extract the communication community-level interest;

[0010] Step S40: introducing an interest combination mechanism based on multi-dimensional knowledge attributes, combining the communication object level interest, the communication community level interest and the communication object importance sequence to obtain a global feature representation of the communication user;

[0011] Step S50: Based on step S40, the local feature representation of the communication user is added to form a communication user feature representation to achieve prediction of the user's call object.

[0012] Preferably, the specific steps of step S10 are as follows:

[0013] Step S101: The user's historical communication record sequence containing multi-dimensional knowledge attributes is represented as O = [o1, o2, ..., o τ ,…,o m ], where m represents the sequence length, o τ Indicates the τth communication record in the user's historical communication records. Each communication record can be represented by a six-tuple o τ =[i τ , c τ , r τ , l τ , t τ , s τ ], where i τ Indicates the number of the communication object in this communication record, c τ Indicates the community to which the communication object in this communication record belongs, r τIndicates the global criticality of the communication object in this communication record, l τ Indicates the communication duration of this communication record, t τ Indicates the communication service type of the communication record, s τ Indicates the calling and called party relationship of the communication record;

[0014] Step S102: construct communication object sequences from historical communication sequences Correspondence Society Sequence Use communication service type data and calling and called type data to construct communication information operation sequence where b τ =pair(t τ , s τ ), pair is a combination function, which maps the combination of different communication service types t and caller and caller types s to the communication information operation type. For example, pair(1, calling) = 1, which means that the communication service type is 1, and the communication information operation type corresponding to the communication record with the caller and caller relationship as "caller" is 1. Similarly, the meanings of pair(1, called) = 2, pair(2, calling) = 3, pair(2, called) = 4, etc. are obtained. The importance sequence is constructed using the communication duration data and the global criticality data. where k τ =Norm(l τ )+Norm(r τ ), Norm function represents the normalization function;

[0015] Preferably, the specific steps of step S20 are as follows:

[0016] Step S201: Initialize the communication object dictionary, communication community dictionary, and communication information operation dictionary with a variance of 0.01 and a mean of 0, and obtain the high-dimensional representations DIC of the communication object dictionary, communication community dictionary, and communication information operation dictionary respectively. I , DIC C , DIC B , respectively extract the high-dimensional space representation of the communication object sequence High-dimensional space representation of communication community sequences High-dimensional space representation of communication information operation sequences represents the high-dimensional representation of the kth communication object in the communication object sequence, represents the high-dimensional space representation of the kth communication community in the communication community sequence, The high-dimensional space representation of the kth communication information operation in the communication information operation sequence, 1≤k≤m, is an m×d-dimensional matrix, where d is the dimension of the high-dimensional space representation;

[0017] In step S202, three gated recurrent unit networks are used to model the communication object, communication community, and communication information operation, and the high-dimensional space representation of the corresponding sequence is updated. The calculation formula is as follows:

[0018]

[0019] Among them GRU I , GRU C , GRU B They represent the gated recurrent unit functions applied to the communication object sequence, communication community sequence, and communication information operation sequence, respectively. are the high-dimensional space representations of the updated k-th communication object, communication community, and communication information operation respectively;

[0020] Preferably, the specific steps of step S30 are as follows:

[0021] Step S301: Represent the communication object sequence in high-dimensional space High-dimensional space representation of sequences of communication information operations Splicing to obtain the communication object-information operation sequence Concat is a vector column-wise concatenation function. Represents the concatenation of the kth column vector in the communication object-information operation representation sequence;

[0022] Step S302: Apply the soft attention mechanism to calculate the difference between each communication object-information operation representation and the last communication object-information operation representation in the communication object-information operation sequence. The similarity between them is combined into the communication object level interest degree, and the calculation formula is as follows:

[0023]

[0024] in, σ represents the Sigmoid activation function;

[0025] Step S303: Represent the communication community sequence in high-dimensional space High-dimensional space representation of sequences of communication information operations Splicing to obtain object community-information operation sequence Represents the object community-information operation representation of the k-th column vector concatenation representation in the sequence;

[0026] Step S304: Apply the soft attention mechanism to extract the representation of each object community-information operation and the last object community-information operation in the object community-information operation representation sequence. The similarity between them is combined into the communication community level interest, and the calculation formula is as follows:

[0027]

[0028] in,

[0029] Preferably, the specific steps of step S40 are as follows:

[0030] Step S401: introduce an interest combination mechanism based on multi-dimensional knowledge attributes, combine the communication object level interest, the communication community level interest and the communication object importance sequence, and calculate the communication user's global interest. The calculation formula is as follows:

[0031] α=w ib ×α ib +w cb ×α cb +w k ×K

[0032] in

[0033] Step S402: Based on the calculated global interest of the communication user, the updated high-dimensional space representation of the communication object is used. h , calculate the global feature representation s of the communication user g , the calculation formula is as follows:

[0034]

[0035] where α k represents the kth interest degree in the global interest degree α of the communication user, High-dimensional space representation of communication object sequence I h The high-dimensional space representation of the k-th communication object in .

[0036] Preferably, the specific steps of step S50 are as follows:

[0037] Step S501: Get the high-dimensional space representation I of the communication object sequence h The last item The local feature representation s of the communication user t , and the global feature representation is concatenated to obtain the communication user feature representation s h =W h [s t ;s g ],in [;] represents splicing, which can predict the communication object at the next moment. The calculation formula is as follows:

[0038]

[0039] in It is the result of sorting the possible communication objects according to probability during the next user communication. The communication object corresponding to the maximum probability is selected as the next communication object of the communication user. Softmax is the activation function, and T represents the matrix transpose.

[0040] The present invention also provides a communication object prediction system based on multi-dimensional knowledge attributes:

[0041] User historical communication record sequence module: constructs a sequence based on the user information in the user historical communication record to obtain the corresponding sequence information, thereby obtaining the user historical communication record sequence, as follows:

[0042] Construct a corresponding communication object sequence according to the communication object number, and construct a communication object dictionary;

[0043] Construct a corresponding communication community sequence according to the community to which the communication object belongs, and construct a communication community dictionary;

[0044] Using a combination function to map the communication service type and the communication caller-caller relationship information into communication information operations, and thereby construct a corresponding communication information operation sequence and a communication information operation dictionary;

[0045] Construct a communication object importance sequence based on the global criticality of the communication object and the communication duration;

[0046] High-dimensional space representation module: Initializes all dictionaries separately to form high-dimensional space representations of corresponding sequences. Using three gated recurrent unit neural networks, it models the communication partner number, communication partner community, and communication information operation, respectively, and updates the high-dimensional space representations of the communication partner sequence, communication community sequence, and communication information operation sequence.

[0047] Hierarchical Interest Module: This module concatenates the high-dimensional spatial representation of the communication object sequence with the high-dimensional spatial representation of the communication information operation sequence to obtain the communication object-information operation sequence, and uses the attention mechanism to extract the hierarchical interest of the communication object from it. It also concatenates the high-dimensional spatial representation of the communication community sequence with the high-dimensional spatial representation of the communication information operation sequence, and then extracts the hierarchical interest of the communication community.

[0048] Global feature representation module: Introduces an interest combination mechanism based on multi-dimensional knowledge attributes, combines the communication object-level interest, the communication community-level interest, and the communication object importance sequence to obtain the global feature representation of the communication user;

[0049] Prediction module: The tail term in the high-dimensional space representation of the communication object sequence is taken as the local feature representation of the communication user. The local feature representation of the communication user is added to the global feature representation of the communication user to form a communication user feature representation, realizing the prediction of the user's call object.

[0050] The beneficial effects of the present invention are as follows: the method of the present invention introduces an interest combination mechanism based on multidimensional knowledge attributes by utilizing the multidimensional knowledge attributes in the user's historical communication records, and analyzes the interest and importance of each communication record in the historical communication sequence to the communication user from three perspectives: communication object, communication community, and communication importance, so as to obtain a global feature representation of the communication user with richer information content, thereby solving the problems of single modeling of the user communication process and insufficient information utilization. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 The figure is a flow chart of the steps of the method of the present invention. DETAILED DESCRIPTION

[0052] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0053] See also Figure 1 , the present invention provides a technical solution:

[0054] The present invention provides the following technical solution: a communication target prediction method and system based on multi-dimensional knowledge attributes, wherein the communication target prediction method comprises the following specific steps:

[0055] Step 1: Based on the user's communication object number, communication object community, communication service type, communication caller relationship, communication object global criticality and communication duration in the user's historical communication records, the corresponding communication object sequence, communication community sequence, communication information operation sequence and communication object importance sequence are constructed, and the corresponding dictionary is constructed;

[0056] Step 1-1: The user's historical communication record sequence containing multi-dimensional knowledge attributes is represented as O = [o1, o2, ..., o τ ,…,o m ], where m represents the sequence length, o τ Indicates the τth communication record in the user's historical communication records. Each communication record can be represented by a six-tuple o τ =[i τ , c τ , rτ , l τ , t τ , s τ ], where i τ Indicates the number of the communication object in this communication record, c τ Indicates the community to which the communication object in this communication record belongs, r τ Indicates the global criticality of the communication object in this communication record, l τ Indicates the communication duration of this communication record, t τ Indicates the communication service type of the communication record, s τ Indicates the calling and called party relationship of the communication record;

[0057] Step 1-2: Construct communication object sequences from historical communication sequences Correspondence Society Sequence Use communication service type data and calling and called type data to construct communication information operation sequence where b τ =pair(t τ , s τ ), pair is a combination function, which maps the combination of different communication service types t and caller and caller types s to the communication information operation type. For example, pair(1, calling) = 1, which means that the communication service type is 1, and the communication information operation type corresponding to the communication record with the caller and caller relationship as "caller" is 1. Similarly, the meanings of pair(1, called) = 2, pair(2, calling) = 3, pair(2, called) = 4, etc. are obtained. The importance sequence is constructed using the communication duration data and the global criticality data. where k τ =Norm(l τ )+Norm(r τ ), Norm function represents the normalization function;

[0058] Step 2: Initialize all dictionaries separately to form high-dimensional spatial representations of the corresponding sequences. Use three gated recurrent unit neural networks to model the communication object number, communication object community, and communication information operation, respectively, and update the high-dimensional spatial representations of the communication object sequence, communication community sequence, and communication information operation sequence.

[0059] Step 2-1: Initialize the communication object dictionary, communication community dictionary, and communication information operation dictionary with a variance of 0.01 and a mean of 0, and obtain the high-dimensional representations DIC of the communication object dictionary, communication community dictionary, and communication information operation dictionary respectively. I , DIC C , DIC B , respectively extract the high-dimensional space representation of the communication object sequence High-dimensional space representation of communication community sequences High-dimensional space representation of communication information operation sequences represents the high-dimensional representation of the kth communication object in the communication object sequence, represents the high-dimensional space representation of the kth communication community in the communication community sequence, The high-dimensional space representation of the kth communication information operation in the communication information operation sequence, 1≤k≤m, is an m×d-dimensional matrix, where d is the dimension of the high-dimensional space representation;

[0060] In step 2-2, three gated recurrent unit networks are used to model the communication object, communication community, and communication information operation, and the high-dimensional space representation of the corresponding sequence is updated. The calculation formula is as follows:

[0061]

[0062] Among them GRU I , GRU C , GRU B They represent the gated recurrent unit functions applied to the communication object sequence, communication community sequence, and communication information operation sequence, respectively. are the high-dimensional space representations of the updated k-th communication object, communication community, and communication information operation respectively;

[0063] Step 3: Concatenate the high-dimensional spatial representation of the communication object sequence with the high-dimensional spatial representation of the communication information operation sequence to obtain the communication object-information operation sequence, and use the attention mechanism to extract the communication object-level interest from it; use a similar method to concatenate the high-dimensional spatial representation of the communication community sequence with the high-dimensional spatial representation of the communication information operation sequence, and then extract the communication community-level interest;

[0064] Step 3-1: Represent the communication object sequence in high-dimensional space High-dimensional space representation of sequences of communication information operations Splicing to obtain the communication object-information operation sequence Concat is a vector column-wise concatenation function. Represents the concatenation of the kth column vector in the communication object-information operation representation sequence;

[0065] Step 3-2: Apply the soft attention mechanism to calculate the representation of each communication object-information operation in the communication object-information operation sequence and the representation of the last communication object-information operation The similarity between them is combined into the communication object level interest degree, and the calculation formula is as follows:

[0066]

[0067] in, σ represents the Sigmoid activation function;

[0068] Step 3-3, represent the communication community sequence in high-dimensional space High-dimensional space representation of sequences of communication information operations Splicing to obtain object community-information operation sequence Represents the object community-information operation representation of the k-th column vector concatenation representation in the sequence;

[0069] Step 3-4: Apply the soft attention mechanism to extract the object community-information operation representation of each object community-information operation representation and the last object community-information operation representation in the object community-information operation representation sequence The similarity between them is combined into the communication community level interest, and the calculation formula is as follows:

[0070]

[0071] in,

[0072] Step 4: Introduce an interest combination mechanism based on multi-dimensional knowledge attributes, combine the communication object-level interest, the communication community-level interest, and the communication object importance sequence to obtain the global feature representation of the communication user;

[0073] Step 4-1: Introduce an interest combination mechanism based on multi-dimensional knowledge attributes, combine the communication object-level interest, the communication community-level interest, and the communication object importance sequence to calculate the communication user's global interest. The calculation formula is as follows:

[0074] α=w ib ×α ib +w cb ×α cb +w k ×K

[0075] in

[0076] Step 4-2: Based on the calculated global interest of the communication user, the updated high-dimensional space representation of the communication object I is used. h , calculate the global feature representation s of the communication user g , the calculation formula is as follows:

[0077]

[0078] where α k represents the kth interest degree in the global interest degree α of the communication user, High-dimensional space representation of communication object sequence I h The high-dimensional space representation of the k-th communication object in .

[0079] Step 5: Based on step S40, the local feature representation of the communication user is added to form a communication user feature representation to achieve prediction of the user's call object.

[0080] Step 5-1, take the high-dimensional space representation I of the communication object sequence h The last item The local feature representation s of the communication user t , and the global feature representation is concatenated to obtain the communication user feature representation s h =W h [s t ;s g ],in [;] represents splicing, which can predict the communication object at the next moment. The calculation formula is as follows:

[0081]

[0082] in It is the result of sorting the possible communication objects according to probability during the next user communication. The communication object corresponding to the maximum probability is selected as the next communication object of the communication user. Softmax is the activation function, and T represents the matrix transpose.

[0083] Although the present invention has been described in detail with reference to the aforementioned embodiments, it is still possible for those skilled in the art to modify the technical solutions described in the aforementioned embodiments, or to make equivalent substitutions for some of the technical features therein. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A communication object prediction method based on multi-dimensional knowledge attributes, characterized in that: The following steps are involved: Step S10: constructing a sequence based on the user information in the user's historical communication record to obtain corresponding sequence information, thereby obtaining a user's historical communication record sequence, as follows: Construct a corresponding communication object sequence according to the communication object number, and construct a communication object dictionary; Construct a corresponding communication community sequence according to the community to which the communication object belongs, and construct a communication community dictionary; Using a combination function to map the communication service type and the communication caller-caller relationship information into communication information operations, and thereby construct a corresponding communication information operation sequence and a communication information operation dictionary; Construct a communication object importance sequence based on the global criticality of the communication object and the communication duration; Step S20: Initialize all dictionaries separately, form high-dimensional spatial representations of corresponding sequences, use three gated recurrent unit neural networks to model the communication object number, communication object community, and communication information operation, and update the high-dimensional spatial representations of the communication object sequence, communication community sequence, and communication information operation sequence; Step S30: Concatenate the high-dimensional spatial representation of the communication object sequence with the high-dimensional spatial representation of the communication information operation sequence to obtain a communication object-information operation sequence, and extract the communication object-level interest from it using the attention mechanism; concatenate the high-dimensional spatial representation of the communication community sequence with the high-dimensional spatial representation of the communication information operation sequence, and then extract the communication community-level interest; Step S40: introducing an interest combination mechanism based on multi-dimensional knowledge attributes, combining the communication object level interest, the communication community level interest and the communication object importance sequence to obtain a global feature representation of the communication user; Step S50: Take the tail item in the high-dimensional space representation of the communication object sequence as the local feature representation of the communication user, and add the local feature representation of the communication user on the basis of step S40 to form a communication user feature representation to achieve prediction of the user's call object.

2. The communication target prediction method based on multi-dimensional knowledge attributes according to claim 1, characterized in that: The specific steps of step S10 are as follows: Step S101: The user's historical communication record sequence containing multi-dimensional knowledge attributes is represented as O = [o1, o2, ..., o τ ,…,o m ], where m represents the sequence length, o τ Indicates the τth communication record in the user's historical communication records. Each communication record can be represented by a six-tuple o τ =[i τ , c τ , r τ , l τ , t τ , s τ ], where i τ Indicates the number of the communication object in this communication record, c τ Indicates the community to which the communication object in this communication record belongs, r τ Indicates the global criticality of the communication object in this communication record, l τ Indicates the communication duration of this communication record, t τ Indicates the communication service type of the communication record, s τ Indicates the calling and called party relationship of the communication record; Step S102: construct communication object sequences from the historical communication record sequence. Correspondence Society Sequence Use communication service type data and calling and called type data to construct communication information operation sequence where b τ =pair(t τ , s τ ), pair is a combination function that maps the combination of different communication service types t and caller / caller types s to the communication information operation type. pair(1, calling) = 1, which means that the communication service type is 1 and the communication record with the caller / caller relationship as "caller" corresponds to the communication information operation type 1. Similarly, the meanings of pair(1, called) = 2, pair(2, calling) = 3, and pair(2, called) = 4 can be obtained. The communication object importance sequence is constructed using communication duration data and global criticality data. where k τ =Norm(l τ )+Norm(r τ ), Norm function represents the normalization function.

3. The communication target prediction method based on multi-dimensional knowledge attributes according to claim 1, characterized in that: The specific steps of step S20 are as follows: Step S201: Initialize the communication object dictionary, communication community dictionary, and communication information operation dictionary with a variance of 0.01 and a mean of 0, and obtain the high-dimensional representations DIC of the communication object dictionary, communication community dictionary, and communication information operation dictionary respectively. I , DIC C , DIC B , respectively extract the high-dimensional space representation of the communication object sequence High-dimensional space representation of communication community sequences High-dimensional space representation of communication information operation sequences represents the high-dimensional representation of the kth communication object in the communication object sequence, represents the high-dimensional space representation of the kth communication community in the communication community sequence, The high-dimensional space representation of the kth communication information operation in the communication information operation sequence, 1≤k≤m, is an m×d-dimensional matrix, where d is the dimension of the high-dimensional space representation; In step S202, three gated recurrent unit networks are used to model the communication object, communication community, and communication information operation, and the high-dimensional space representation of the corresponding sequence is updated. The calculation formula is as follows: Among them GRU I , GRU C , GRU B Represent the gated recurrent unit functions applied to the communication object sequence, communication community sequence, and communication information operation sequence, respectively. are the high-dimensional space representations of the updated k-th communication object, communication community, and communication information operation, respectively.

4. The communication target prediction method based on multi-dimensional knowledge attributes according to claim 1, characterized in that: The specific steps of step S30 are as follows: Step S301: Represent the communication object sequence in high-dimensional space High-dimensional space representation of sequences of communication information operations Splicing to obtain the communication object-information operation sequence Concat is a vector column-wise concatenation function. Represents the concatenation of the kth column vector in the communication object-information operation representation sequence; Step S302: Apply the soft attention mechanism to calculate the difference between each communication object-information operation representation and the last communication object-information operation representation in the communication object-information operation sequence. The similarity between them is combined into the communication object level interest degree, and the calculation formula is as follows: in, σ represents the Sigmoid activation function; Step S303: Represent the communication community sequence in high-dimensional space High-dimensional space representation of sequences of communication information operations Splicing to obtain object community-information operation sequence Represents the object community-information operation representation of the k-th column vector concatenation representation in the sequence; Step S304: Apply the soft attention mechanism to extract the information operation representation of each object community and the last object community in the object community-information operation representation sequence. The similarity between them is combined into the communication community level interest, and the calculation formula is as follows: in, 5. The communication target prediction method based on multi-dimensional knowledge attributes according to claim 1, characterized in that: The specific steps of step S40 are as follows: Step S401: introduce an interest combination mechanism based on multi-dimensional knowledge attributes, combine the communication object level interest, the communication community level interest and the communication object importance sequence, and calculate the communication user's global interest. The calculation formula is as follows: α=w ib ×α ib +w cb ×α cb +w k ×K in w ib , w cb , w k are the weights corresponding to interest and importance, respectively, which are set as trainable parameters; Step S402: Based on the calculated global interest of the communication user, the updated high-dimensional space representation of the communication object is used. h , calculate the global feature representation s of the communication user g , the calculation formula is as follows: where α k represents the kth interest degree in the global interest degree α of the communication user, High-dimensional space representation of communication object sequence I h The high-dimensional space representation of the k-th communication object in .

6. The communication target prediction method based on multi-dimensional knowledge attributes according to claim 1, characterized in that: The specific steps of step S50 are as follows: Step S501: Get the high-dimensional space representation I of the communication object sequence h The last item The local feature representation s of the communication user t , and the global feature representation is concatenated to obtain the communication user feature representation s h =W h [s t ;s g ],in [;] represents splicing, which can predict the communication object at the next moment. The calculation formula is as follows: in It is the result of sorting the possible communication objects according to probability during the next user communication. The communication object corresponding to the maximum probability is selected as the next communication object of the communication user. Softmax is the activation function, and T represents the matrix transpose.

7. A communication target prediction system based on multi-dimensional knowledge attributes, characterized by: User historical communication record sequence module: constructs a sequence based on the user information in the user historical communication record to obtain the corresponding sequence information, thereby obtaining the user historical communication record sequence, as follows: Construct a corresponding communication object sequence according to the communication object number, and construct a communication object dictionary; Construct a corresponding communication community sequence according to the community to which the communication object belongs, and construct a communication community dictionary; Using a combination function to map the communication service type and the communication caller-caller relationship information into communication information operations, and thereby construct a corresponding communication information operation sequence and a communication information operation dictionary; Construct a communication object importance sequence based on the global criticality of the communication object and the communication duration; High-dimensional space representation module: Initializes all dictionaries separately to form high-dimensional space representations of corresponding sequences. Using three gated recurrent unit neural networks, it models the communication partner number, communication partner community, and communication information operation, respectively, and updates the high-dimensional space representations of the communication partner sequence, communication community sequence, and communication information operation sequence. Hierarchical Interest Module: This module concatenates the high-dimensional spatial representation of the communication object sequence with the high-dimensional spatial representation of the communication information operation sequence to obtain the communication object-information operation sequence, and uses the attention mechanism to extract the hierarchical interest of the communication object from it. It also concatenates the high-dimensional spatial representation of the communication community sequence with the high-dimensional spatial representation of the communication information operation sequence, and then extracts the hierarchical interest of the communication community. Global feature representation module: Introduces an interest combination mechanism based on multi-dimensional knowledge attributes, combines the communication object-level interest, the communication community-level interest, and the communication object importance sequence to obtain the global feature representation of the communication user; Prediction module: The tail term in the high-dimensional space representation of the communication object sequence is taken as the local feature representation of the communication user. The local feature representation of the communication user is added to the global feature representation of the communication user to form a communication user feature representation, realizing the prediction of the user's call object.

Images