A monitoring method, system and storage medium for automobile marketing data

Abstract

The present invention relates to the field of automobile marketing technology, and discloses a monitoring method, system and storage medium for automobile marketing data, comprising the following steps: collecting text information of a current model and historically sold models under different marketing plans; performing NER on the text information of the current model to obtain named entities; matching the named entities extracted in the above steps with entities in an automobile knowledge graph and marking the matched entities; calculating the similarity between automobile parameter features and marketing features of different entities of the current model; selecting the marketing feature with the greatest similarity as the i-th recommended marketing plan, deleting duplicate items in the obtained set of recommended marketing plans, and obtaining a set of marketing plans as the final recommended marketing plan; the present invention focuses on obtaining various automobile parameter features of a model and marketing features of a marketing plan through a pre-trained language model, thereby improving the matching accuracy between the marketing plan and the current model.

Description

Technical Field

[0001] The present invention relates to the field of automobile marketing technology, and more specifically, to a monitoring method, system and storage medium for automobile marketing data. Background Art

[0002] Traditional marketing intelligence recommendation systems usually judge the advantages of the current model based on the parameter information of the current model, so as to recommend a suitable marketing plan for the current model. However, such systems have limitations when multiple marketing plans need to be recommended for reference for the current model, especially when the parameter information of the current model contains multiple different car parameters at the same time. It is impossible to recommend suitable marketing plans for the different car parameters of the current model. Summary of the invention

[0003] The present invention provides a monitoring method, system and storage medium for automobile marketing data, which obtains different entities existing in the text information of the current model, expands the text content of the entity through the automobile knowledge graph, extracts features from the text, calculates the similarity between the automobile parameter features of each entity and the marketing features of each marketing plan for different entities, thereby recommending a marketing plan for different automobile parameters existing in the current model, solving the technical problem that it is impossible to recommend a suitable marketing plan for multiple automobile parameters of the current model.

[0004] The present invention provides a method for monitoring automobile marketing data, comprising the following steps:

[0005] Step 100, collecting text information of the current model and historically sold models under different marketing plans, the text information including target audience, appearance and interior, power and safety, intelligence and endurance, and price;

[0006] Step 200, performing NER named entity recognition on the text information of the current vehicle model to obtain a named entity;

[0007] Step 300, matching the named entities extracted in step 200 with entities in the automobile knowledge graph and marking the matched entities, and performing m non-repetitive random walks with the marked entities as the center, where the step length of the random walk is 1, the number of steps is n, the default value of n is 10, and the default value of m is 10; sorting the entities of each random walk according to the order of the random walk to obtain the second text;

[0008] Step 400, extracting a set of text features from the second text obtained by random walking of the same marked entity through BERT text feature extraction, averaging the set of text features to obtain automobile parameter features, and sequentially obtaining automobile parameter features of entities with different marks;

[0009] The calculation formula is as follows:

[0010] ;

[0011] in, represents the car parameter features obtained after a random walk of each labeled entity, Represents the averaged car parameter features of each labeled entity;

[0012] Step 500, extracting text information of historically sold models under different marketing plans through BERT text feature extraction to obtain text features of historically sold models under different marketing plans; calculating marketing features;

[0013] The marketing characteristics calculation formula is as follows:

[0014] ;

[0015] in Indicates that the y-th marketing plan is equivalent to the marketing characteristics of the i-th labeled entity, represents the set of historically sold models of the y-th marketing plan, and d represents The total number of models, express The car text features of the p-th car model in, represents the car model parameter features of the i-th marked entity, represents the similarity function;

[0016] Step 600, calculating the similarity between the automobile parameter characteristics of the i-th marked entity and the marketing characteristics of each marketing plan relative to the i-th marked entity; selecting the marketing plan with the greatest similarity as the i-th recommended marketing plan;

[0017] The similarity calculation formula is as follows:

[0018] ;

[0019] in, represents the similarity value between the car parameter features of the ith labeled entity and the marketing features of the yth marketing plan relative to the ith labeled entity, represents the similarity function;

[0020] Step 700: Deleting duplicate items from the obtained set of recommended marketing solutions to obtain a final set of recommended marketing solutions.

[0021] Further, step 200 includes the following steps:

[0022] Step 201, converting text information of the current vehicle model into a vocabulary sequence;

[0023] Step 202, using the embedding layer pre-trained by the BERT model to convert the vocabulary sequence into an embedding vector sequence;

[0024] Step 203, passing the embedding vector sequence into the BERT model to obtain the output of the BERT model;

[0025] Step 204, the output of the BERT model is passed to the CRF layer, and the CRF layer performs sequence labeling to obtain a label sequence, where the rth label of the label sequence represents the entity type of the rth word in the vocabulary sequence.

[0026] Step 205 : extracting, according to the predicted tag sequence, the words corresponding to the tags belonging to the named entity in the tag sequence as the named entity.

[0027] Furthermore, the Transformer encoder calculation formula of the BERT model in step 203 is as follows:

[0028] Calculate the Query matrix of the i-th head of the Transformer encoder 、Key Matrix and Value matrix :

[0029] ;

[0030] in , , are the Query weight matrix, Key weight matrix, and Value weight matrix of the i-th head of the Transformer encoder, is the embedding vector of the input;

[0031] Calculate the attention weight matrix A:

[0032] ;

[0033] The weight matrix A and the Value matrix Multiply to get the output of the i-th head of the Transformer encoder, softmax represents the normalized exponential function, represents the dimension of the embedding vector, and T represents the transpose:

[0034] ;

[0035] ;

[0036] in, is the output layer weight matrix, represents the output of the i-th head of the Transformer encoder, , , Represent the outputs of the 1st, 2nd, and Rth heads of the Transformer encoder, respectively. Represents the output of the BERT model.

[0037] Furthermore, using the automobile knowledge graph to match entities to obtain the second text further includes the following steps:

[0038] Step 401, entity matching, the entities extracted in step 205 are grouped into a set , where each entity is represented by e; for the set Each entity e in is matched by calculating the cosine similarity with the entities in the automobile knowledge graph;

[0039] Step 402, random walk, for each matched entity, m random walks are performed, and the step length of each walk is n steps;

[0040] Step 403, entity sequence generation, arrange the results of each random walk in order to obtain a second text.

[0041] Furthermore, the automobile knowledge graph includes the following contents:

[0042] The automobile knowledge graph includes entities representing the following information: model, appearance, interior, price, engine, intelligent driving, and marketing plan;

[0043] The automobile knowledge graph expresses the relationship between information such as vehicle models, parameters, and marketing plans, among which the parameters include appearance, interior, engine, price, and intelligent driving.

[0044] Furthermore, a random walk is performed on the automobile knowledge graph with the named entity of the final recommended marketing plan as the center to obtain an entity set, which is named H. The entities in an entity set H include model, appearance, interior, and price.

[0045] Graph structure data is established based on the set of marketing plans recommended at the end, the entity set H, and the automobile parameter features of the marked entities. The graph structure data includes nodes and edges connecting the nodes. A node represents a marketing plan recommended at the end, an entity in the entity set H, or an automobile parameter feature of a marked entity.

[0046] The edges between the nodes representing the entities in the entity set H or the final recommended marketing solutions represent the entity relationships between the entities in the entity set H or the final recommended marketing solutions on the automobile knowledge graph;

[0047] The two nodes represent the automobile parameter features of the marked entity and the final recommended marketing plan respectively. If the similarity between the parameter features of the marked entity and the final recommended marketing plan relative to the marketing features of the marked entity is the largest, then there is an edge between the two nodes, otherwise there is no edge;

[0048] The graph structure data is input into the trained graph neural network, which includes a hidden layer and an output layer.

[0049] Furthermore, the calculation formula of the hidden layer of the graph neural network is as follows:

[0050] ;

[0051] in represents the encoding feature of the vth node, is the set of neighbors of node v, represents the input features of node u, is a normalizing constant, is a nonlinear activation function, W represents the weight parameter of the hidden layer, and b represents the bias parameter of the hidden layer;

[0052] The input features of the node representing the marked entity are the car parameter features of the marked entity, the input features of the node representing the entity in the entity set H are obtained by encoding the entity information of the entity, and the input features of the node representing the entity of the marketing plan are obtained by encoding the text information of the marketing plan;

[0053] The entity information of an entity in set H represents the information associated with the entity. If an entity is an interior, then its associated entity information is the description text corresponding to the interior.

[0054] The output layer inputs the encoding features of the nodes representing the final recommended marketing plan, and outputs the value of the serial number representing the implementation order of the final recommended marketing plan.

[0055] Furthermore, the input feature of the node representing the marked entity is the automobile parameter feature of the marked entity, the input feature of the node representing the entity in the entity set H is obtained by encoding the entity information of the entity, and the input feature of the node representing the entity of the marketing plan is obtained by encoding the text information of the marketing plan;

[0056] The entity information of an entity in set H represents the information associated with the entity. If an entity is an interior, then its associated entity information is the description text corresponding to the interior.

[0057] The output layer inputs the encoding features of the nodes representing the final recommended marketing plan, and outputs the value of the serial number representing the implementation order of the final recommended marketing plan.

[0058] The present invention provides a monitoring system for automobile marketing data, which is used to execute the aforementioned monitoring method for automobile marketing data, including:

[0059] The entity recognition module obtains named entities by performing NER named entity recognition on the text information of the current vehicle model;

[0060] The entity matching expansion module matches the extracted named entities with the entities in the automobile knowledge graph, performs random walks on the matched entities, and sorts the entities in each random walk according to the order of the random walk to obtain the second text;

[0061] The feature encoding module extracts a set of text features from the second text obtained by random walking of the same entity through BERT text features, and obtains the vehicle parameter features by averaging. At the same time, the text information of historical vehicle models under different marketing plans is extracted through BERT text features to obtain the text features of historical vehicle models under different marketing plans.

[0062] The calculation formula is as follows:

[0063] ;

[0064] in, represents the car parameter features obtained after a random walk of each labeled entity, Represents the averaged car parameter features of each labeled entity;

[0065] The marketing plan matching module calculates the cosine similarity value of the features of the current model through the vehicle parameter features and the marketing features calculated from the historical model text features under different marketing plans;

[0066] The marketing characteristics calculation formula is as follows:

[0067] ;

[0068] in Indicates that the y-th marketing plan is equivalent to the marketing characteristics of the i-th labeled entity, represents the set of historically sold models of the y-th marketing plan, and d represents The total number of models, express The car text features of the p-th car model in, represents the car model parameter features of the i-th marked entity, represents the similarity function;

[0069] The similarity calculation formula is as follows:

[0070] ;

[0071] in, represents the similarity value between the car parameter features of the ith labeled entity and the marketing features of the yth marketing plan relative to the ith labeled entity, represents the similarity function;

[0072] The marketing plan recommendation module removes duplicate marketing plans and obtains a final set of recommended marketing plans;

[0073] The cost accounting module calculates the total cost of all plans based on the costs involved in the marketing plan;

[0074] The marketing program scoring module is used to record the final effect score of the marketing program used by the current model and to calculate the score, and the statistical result is the average score.

[0075] The present invention provides a storage medium storing non-transitory computer-readable instructions. When the non-transitory computer-readable instructions are executed by a computer, the steps in the above-mentioned method for monitoring automobile marketing data can be executed.

[0076] The beneficial effects of the present invention are:

[0077] 1. The present invention can identify in detail the different automobile parameter information existing in the current vehicle model text through named entity recognition and knowledge graph technology, thereby obtaining more fine-grained automobile parameter features of the current vehicle model;

[0078] 2. The present invention calculates the different automobile parameter characteristics of the current model and the automobile parameter characteristics of the historical models under different marketing plans through the attention mechanism to obtain the marketing characteristics of the marketing plan for the different automobile parameters of the current model, thereby improving the pertinence of the marketing characteristics. BRIEF DESCRIPTION OF THE DRAWINGS

[0079] Figure 1 is a flow chart of a monitoring method for automobile marketing data of the present invention;

[0080] Figure 2 It is a flowchart of named entity recognition in a monitoring method for automobile marketing data of the present invention;

[0081] Figure 3 It is a module schematic diagram of a monitoring system for automobile marketing data of the present invention. DETAILED DESCRIPTION

[0082] The subject matter described herein will now be discussed with reference to example implementations. It should be understood that the discussion of these implementations is only to enable those skilled in the art to better understand and implement the subject matter described herein, and the functions and arrangements of the elements discussed may be changed without departing from the scope of protection of the present specification. Various examples may omit, replace, or add various processes or components as needed. In addition, the features described in some examples may also be combined in other examples.

[0083] like Figure 1 As shown, a monitoring method for automobile marketing data includes the following steps:

[0084] Step 100, collecting text information of the current model and historically sold models under different marketing plans, the text information including target audience, appearance and interior, power and safety, intelligence and endurance, and price;

[0085] Step 200, performing NER named entity recognition on the text information of the current vehicle model to obtain a named entity;

[0086] In one embodiment of the present invention, the NER named entity recognition in step 200 provides an example of obtaining the encoding:

[0087] Convert the text information of the current car model into a vocabulary sequence as follows:

[0088] X=(compact, SUV, length 4.51m, width 1.84m, height 1.68m, leather interior, 1.5T, engine, 132kW, 300N·m, 6 airbags, L2 intelligent driving, 109,800);

[0089] Step 300, matching the named entities extracted in step 200 with entities in the automobile knowledge graph and marking the matched entities, and performing m non-repetitive random walks with the marked entities as the center, where the step length of the random walk is 1, the number of steps is n, the default value of n is 10, and the default value of m is 10; sorting the entities of each random walk according to the order of the random walk to obtain the second text;

[0090] The automobile knowledge graph includes entities representing the following information: vehicle model, interior, appearance, engine, intelligent driving, price, and marketing plan.

[0091] The automobile knowledge graph expresses the correlation between information such as vehicle models, parameters, and marketing plans. The parameters include appearance, interior, engine, intelligent driving and other information other than vehicle models.

[0092] In one embodiment of the present invention, step 300 includes the following steps:

[0093] Entity matching, the matching rule is: if the similarity value between the entity in the automobile knowledge graph and the named entity extracted in step 200 is high, then the entity in the automobile knowledge graph is matched with the named entity extracted in step 200, otherwise it is not matched;

[0094] Random walk: for each matching entity, m random walks are performed, and each walk has a step length of n steps;

[0095] Entity sequence generation: Arrange the results of each random walk in order to obtain the second text S.

[0096] Step 400, extracting text features from the second text S obtained by random walking of the same marked entity to obtain a group of text features, averaging the group of text features to obtain automobile parameter features, and sequentially obtaining automobile parameter features of entities with different marks;

[0097] In one embodiment of the present invention, the calculation formula of the automobile parameter characteristics of the entities with different tags in step 400 is as follows:

[0098] ;

[0099] in, represents the car parameter features obtained after a random walk of each labeled entity, Represents the averaged car parameter features of each labeled entity;

[0100] For example, the obtained car parameter features are , , , it means that the current vehicle model information text contains 3 different vehicle parameters.

[0101] Step 500, extracting text information of historically sold models under different marketing plans through BERT text feature extraction to obtain text features of historically sold models under different marketing plans; calculating marketing features;

[0102] In one embodiment of the present invention, the marketing feature calculation formula in step 500 is as follows:

[0103] ;

[0104] in Indicates that the y-th marketing plan is equivalent to the marketing characteristics of the i-th labeled entity, represents the set of historically sold models of the y-th marketing plan, and d represents The total number of models, express The car text features of the p-th car model in, represents the car parameter features of the model of the i-th labeled entity, Represents the similarity function.

[0105] Step 600, calculating the similarity between the automobile parameter characteristics of the i-th marked entity and the marketing characteristics of each marketing plan relative to the i-th marked entity; selecting the marketing plan with the greatest similarity as the i-th recommended marketing plan;

[0106] In one embodiment of the present invention, the similarity calculation formula in step 600 is as follows:

[0107] ;

[0108] in, represents the similarity value between the car parameter features of the ith labeled entity and the marketing features of the yth marketing plan relative to the ith labeled entity, Represents a similarity function, such as the cosine similarity function.

[0109] Step 700: Deleting duplicate items from the obtained set of recommended marketing solutions to obtain a final set of recommended marketing solutions.

[0110] In one embodiment of the present invention, the set of marketing plans in step 700 is as follows:

[0111] ;

[0112] Among them, G represents the set of recommended marketing plans.

[0113] like Figure 2 As shown, step 200 includes the following steps:

[0114] Step 201, converting text information of the current vehicle model into a vocabulary sequence;

[0115] In one embodiment of the present invention, the text conversion vocabulary in step 201 is as follows:

[0116] X=(compact, SUV, length 4.51m, width 1.84m, height 1.68m, leather interior, 1.5T, engine, 132kW, 300N·m, 6 airbags, L2 intelligent driving, 109,800);

[0117] Step 202, using the embedding layer pre-trained by the BERT (Bidirectional Encoder Representations from Transformers) model to convert the vocabulary sequence into an embedding vector sequence;

[0118] In one embodiment of the present invention, the vocabulary sequence in step 202 is converted into an embedding vector sequence as follows:

[0119] ;

[0120] in They represent the embedding vectors corresponding to the words compact, SUV, length 4.51m, width 1.84m, height 1.68m, leather interior, 1.5T, engine, 132kW, 300N·m, 6 airbags, L2 intelligent driving, and 109,800 respectively.

[0121] Step 203, passing the embedding vector sequence into the BERT model to obtain the output of the BERT model;

[0122] In one embodiment of the present invention, the Transformer encoder calculation formula of the BERT model in step 203 is as follows:

[0123] Calculate the Query matrix of the i-th head of the Transformer encoder 、Key Matrix and Value matrix :

[0124] ;

[0125] in , , are the Query weight matrix, Key weight matrix, and Value weight matrix of the i-th head of the Transformer encoder, is the embedding vector of the input;

[0126] Calculate the attention weight matrix A:

[0127] ;

[0128] The weight matrix A and the Value matrix Multiply to get the output of the i-th head of the Transformer encoder, softmax represents the normalized exponential function, represents the dimension of the embedding vector, and T represents the transpose:

[0129] ;

[0130] ;

[0131] in, is the output layer weight matrix, represents the output of the i-th head of the Transformer encoder, , , Respectively represent the output of the 1st, 2nd, and Rth heads of the Transformer encoder, Represents the output of the BERT model;

[0132] In step 204, the output of the BERT model is passed to the CRF (conditional random field) layer, and the CRF layer performs sequence labeling to obtain a label sequence, where the rth label of the label sequence represents the entity type of the rth word in the vocabulary sequence.

[0133] In one embodiment of the present invention, a tag sequence in step 204 is as follows:

[0134] ;

[0135] The meanings of the tags are as follows:

[0136] O: non-named entity, B-Parameters: named entity;

[0137] Step 205 : extracting, according to the predicted tag sequence, the words corresponding to the tags belonging to the named entity in the tag sequence as the named entity.

[0138] In one embodiment of the present invention, the extraction of named entities in step 205 is represented as follows:

[0139] SUV, length, width 1.84m, height 1.68m, 1.5T, 132kW, 300N·m, 6 airbags, L2 intelligent driving, 109,800.

[0140] In the aforementioned embodiment, an actual phenomenon is that the sequential combination of the marketing plans generated in the above step 700 is random and disordered, which may weaken the synergy between the marketing plans, because a carefully planned schedule can ensure that each marketing activity can play the greatest role and form synergy with each other.

[0141] For example, for some cars, quickly attracting a group of potential customers through online marketing, and then holding offline product exhibitions to further deepen their understanding of the products will achieve better marketing results.

[0142] Since the recommended marketing plans generally need to cross scenarios, the current car model's marketing promotion will generate different marketing costs and time costs depending on the scenario. Reasonable planning of the marketing sequence can reduce the time cost and money cost of the current car model's marketing promotion.

[0143] In one embodiment of the present invention, a method for monitoring automobile marketing data is provided, and the following operations are performed after obtaining a set of final recommended marketing plans:

[0144] On the automobile knowledge graph, a random walk is performed with the named entity of the final recommended marketing plan as the center to obtain an entity set, which is named H. The entities in an entity set H include model, appearance, interior, and price.

[0145] Graph structure data is established based on the set of marketing plans recommended at the end, the entity set H, and the automobile parameter features of the marked entities. The graph structure data includes nodes and edges connecting the nodes. A node represents a marketing plan recommended at the end, an entity in the entity set H, or an automobile parameter feature of a marked entity.

[0146] The edges between the nodes representing the entities in the entity set H or the final recommended marketing solutions represent the entity relationships between the entities in the entity set H or the final recommended marketing solutions on the automobile knowledge graph;

[0147] The two nodes represent the automobile parameter features of the marked entity and the final recommended marketing plan respectively. If the similarity between the parameter features of the marked entity and the final recommended marketing plan relative to the marketing features of the marked entity is the largest, then there is an edge between the two nodes, otherwise there is no edge;

[0148] The similarity between the parameter characteristics of the marked entity and the final recommended marketing plan relative to the marketing characteristics of the marked entity is the greatest if:

[0149] The value set of similarities between the parameter feature of the marked entity and all marketing plans relative to the marketing feature of the marked entity is the largest.

[0150] Input the graph structure data into the trained graph neural network, which includes a hidden layer and an output layer;

[0151] The calculation formula of the hidden layer is as follows:

[0152] ;

[0153] in represents the encoding feature of the vth node, is the set of neighbors of node v, represents the input features of node u, is a normalizing constant (such as node degree), is a nonlinear activation function, b represents the bias parameter of the hidden layer, and W represents the weight parameter of the hidden layer.

[0154] The input features of the node representing the marked entity are the car parameter features of the marked entity, the input features of the node representing the entity in the entity set H are obtained by encoding the entity information of the entity, and the input features of the node representing the entity of the marketing plan are obtained by encoding the text information of the marketing plan.

[0155] The entity information of the entity in the set H represents the information associated with the entity. For example, if an entity is an interior, its associated entity information is the description text corresponding to the interior.

[0156] The output layer inputs the encoding features of the nodes representing the final recommended marketing plan, and outputs the value of the serial number representing the implementation order of the final recommended marketing plan.

[0157] The text information of a marketing plan refers to the text that describes the marketing plan process, ideas, etc.

[0158] For example, if the coding feature of the node of the fourth final recommended marketing plan is input and the output value is 3, it means that the final recommended marketing plan is ranked fourth for implementation.

[0159] like Figure 3 As shown, a monitoring system for automobile marketing data includes the following modules:

[0160] The entity recognition module 301 obtains a named entity by performing NER named entity recognition on the text information of the current vehicle model;

[0161] The entity matching expansion module 302 matches the extracted named entities with entities in the automobile knowledge graph, performs random walks on the matched entities, and sorts the entities of each random walk according to the order of the random walks to obtain a second text;

[0162] The feature encoding module 303 extracts a set of text features from the second text obtained by random walking of the same entity through BERT text features, and obtains the vehicle parameter features by averaging. Meanwhile, the text information of historical vehicle models under different marketing plans is extracted through BERT text features to obtain the text features of historical vehicle models under different marketing plans.

[0163] The marketing scheme matching module 304 calculates the cosine similarity value of the features of the current vehicle model and the marketing features calculated from the vehicle model text features under different marketing schemes;

[0164] The marketing plan recommendation module 305 removes duplicate marketing plans and obtains a set of final recommended marketing plans;

[0165] The cost accounting module 306 calculates the total cost of all plans based on the costs involved in the marketing plans;

[0166] The marketing program scoring module 307 is used to record the final effect score of the marketing program used by the current model and to calculate the score, and the result of the statistics is the average score.

[0167] For example, if the marketing plan is scored on a 100-point scale, and the three current car models score 60, 50, and 90 respectively, the statistical average score is 66.7.

[0168] A storage medium stores non-transitory computer-readable instructions. When the non-transitory computer-readable instructions are executed by a computer, the steps in the above-mentioned method for monitoring automobile marketing data can be executed.

[0169] The above describes an embodiment of the present invention, but this embodiment is not limited to the above-mentioned specific implementation mode. The above-mentioned specific implementation mode is merely illustrative and not restrictive. Under the guidance of this embodiment, ordinary technicians in this field can also make more forms of equivalent embodiments, all of which are within the protection of this embodiment.

Claims

1. A monitoring method for automobile marketing data, characterized in that: The following steps are involved: Step 100, collecting text information of the current model and historically sold models under different marketing plans, the text information including target audience, appearance and interior, power and safety, intelligence and endurance, and price; Step 200, performing NER named entity recognition on the text information of the current vehicle model to obtain a named entity; Step 300, matching the named entities extracted in step 200 with entities in the automobile knowledge graph and marking the matched entities, and performing m non-repetitive random walks with the marked entities as the center, where the step length of the random walk is 1, the number of steps is n, the default value of n is 10, and the default value of m is 10; sorting the entities of each random walk according to the order of the random walk to obtain the second text; Step 400, extracting a set of text features from the second text obtained by random walking of the same marked entity through BERT text feature extraction, averaging the set of text features to obtain automobile parameter features, and sequentially obtaining automobile parameter features of entities with different marks; The calculation formula is as follows: ； in, represents the car parameter features obtained after a random walk of each labeled entity, Represents the averaged car parameter features of each labeled entity; Step 500, extracting text information of historically sold models under different marketing plans through BERT text feature extraction to obtain text features of historically sold models under different marketing plans; calculating marketing features; The marketing characteristics calculation formula is as follows: ； in Indicates that the y-th marketing plan is equivalent to the marketing characteristics of the i-th labeled entity, represents the set of historically sold models of the y-th marketing plan, and d represents The total number of models, express The car text features of the p-th car model in, represents the car model parameter features of the i-th labeled entity, represents the similarity function; Step 600, calculating the similarity between the automobile parameter characteristics of the i-th marked entity and the marketing characteristics of each marketing plan relative to the i-th marked entity; selecting the marketing plan with the greatest similarity as the i-th recommended marketing plan; The similarity calculation formula is as follows: ； in, represents the similarity value between the car parameter features of the ith labeled entity and the marketing features of the yth marketing plan relative to the ith labeled entity, represents the similarity function; Step 700: Deleting duplicate items from the obtained set of recommended marketing solutions to obtain a final set of recommended marketing solutions.

2. A method for monitoring automobile marketing data according to claim 1, characterized in that: Step 200 includes the following steps: Step 201, converting text information of the current vehicle model into a vocabulary sequence; Step 202, using the embedding layer pre-trained by the BERT model to convert the vocabulary sequence into an embedding vector sequence; Step 203, passing the embedding vector sequence into the BERT model to obtain the output of the BERT model; Step 204, the output of the BERT model is passed to the CRF layer, and the CRF layer performs sequence labeling to obtain a label sequence, where the rth label of the label sequence represents the entity type of the rth word in the vocabulary sequence; Step 205 : extracting, according to the predicted tag sequence, the words corresponding to the tags belonging to the named entity in the tag sequence as the named entity.

3. A method for monitoring automobile marketing data according to claim 2, characterized in that: The Transformer encoder calculation formula of the BERT model in step 203 is as follows: Calculate the Query matrix of the i-th head of the Transformer encoder 、Key Matrix and Value matrix : ； in , , are the Query weight matrix, Key weight matrix, and Value weight matrix of the i-th head of the Transformer encoder, is the embedding vector of the input; Calculate the attention weight matrix A: ； The weight matrix A and the Value matrix Multiply to get the output of the i-th head of the Transformer encoder, softmax represents the normalized exponential function, represents the dimension of the embedding vector, and T represents the transpose: ； ； in, is the output layer weight matrix, represents the output of the i-th head of the Transformer encoder, , , Represent the outputs of the 1st, 2nd, and Rth heads of the Transformer encoder, respectively. Represents the output of the BERT model.

4. The method for monitoring automobile marketing data according to claim 1, characterized in that: Using the automobile knowledge graph to match entities to obtain the second text also includes the following steps: Step 401, entity matching, the entities extracted in step 205 are grouped into a set , where each entity is represented by e; for the set Each entity e in is matched by calculating the cosine similarity with the entities in the automobile knowledge graph; Step 402, random walk, for each matched entity, m random walks are performed, and the step length of each walk is n steps; Step 403, entity sequence generation, arrange the results of each random walk in order to obtain a second text.

5. A method for monitoring automobile marketing data according to claim 4, characterized in that: The automobile knowledge graph includes the following: The automobile knowledge graph includes entities representing the following information: model, appearance, interior, price, engine, intelligent driving, and marketing plan; The automobile knowledge graph expresses the relationship between vehicle models, parameters, and marketing plan information, where the parameters include appearance, interior, engine, price, and intelligent driving.

6. A method for monitoring automobile marketing data according to claim 5, characterized in that: On the automobile knowledge graph, a random walk is performed with the named entity of the final recommended marketing plan as the center to obtain an entity set, which is named H. The entities in an entity set H include model, appearance, interior, and price. Graph structure data is established based on the set of marketing plans recommended at the end, the entity set H, and the automobile parameter features of the marked entities. The graph structure data includes nodes and edges connecting the nodes. A node represents a marketing plan recommended at the end, an entity in the entity set H, or an automobile parameter feature of a marked entity. The edges between the nodes representing the entities in the entity set H or the final recommended marketing solutions represent the entity relationships between the entities in the entity set H or the final recommended marketing solutions on the automobile knowledge graph; The two nodes represent the automobile parameter features of the marked entity and the final recommended marketing plan respectively. If the similarity between the parameter features of the marked entity and the final recommended marketing plan relative to the marketing features of the marked entity is the largest, then there is an edge between the two nodes, otherwise there is no edge; The graph structure data is input into the trained graph neural network, which includes a hidden layer and an output layer.

7. A method for monitoring automobile marketing data according to claim 6, characterized in that: The calculation formula of the hidden layer of the graph neural network is as follows: ； in represents the encoding feature of the vth node, is the set of neighbors of node v, represents the input features of node u, is a normalizing constant, is a nonlinear activation function, W represents the weight parameter of the hidden layer, and b represents the bias parameter of the hidden layer; The input features of the node representing the marked entity are the car parameter features of the marked entity, the input features of the node representing the entity in the entity set H are obtained by encoding the entity information of the entity, and the input features of the node representing the entity of the marketing plan are obtained by encoding the text information of the marketing plan; The entity information of an entity in set H represents the information associated with the entity. If an entity is an interior, then its associated entity information is the description text corresponding to the interior. The output layer inputs the encoding features of the nodes representing the final recommended marketing plan, and outputs the value of the serial number representing the implementation order of the final recommended marketing plan.

8. The method for monitoring automobile marketing data according to claim 7, characterized in that: The input features of the node representing the marked entity are the car parameter features of the marked entity, the input features of the node representing the entity in the entity set H are obtained by encoding the entity information of the entity, and the input features of the node representing the entity of the marketing plan are obtained by encoding the text information of the marketing plan; The entity information of an entity in set H represents the information associated with the entity. If an entity is an interior, then its associated entity information is the description text corresponding to the interior. The output layer inputs the encoding features of the nodes representing the final recommended marketing plan, and outputs the value of the serial number representing the implementation order of the final recommended marketing plan.

9. A monitoring system for automobile marketing data, characterized in that: It is used to execute a monitoring method for automobile marketing data as described in any one of claims 1 to 8, comprising: The entity recognition module obtains named entities by performing NER named entity recognition on the text information of the current vehicle model; The entity matching expansion module matches the extracted named entities with the entities in the automobile knowledge graph, performs random walks on the matched entities, and sorts the entities in each random walk according to the order of the random walk to obtain the second text; The feature encoding module extracts a set of text features from the second text obtained by random walking of the same entity through BERT text features, and obtains the vehicle parameter features by averaging. At the same time, the text information of historical vehicle models under different marketing plans is extracted through BERT text features to obtain the text features of historical vehicle models under different marketing plans. The calculation formula is as follows: ； in, represents the car parameter features obtained after a random walk of each labeled entity, Represents the averaged car parameter features of each labeled entity; The marketing plan matching module calculates the cosine similarity value of the features of the current model through the vehicle parameter features and the marketing features calculated from the historical model text features under different marketing plans; The marketing characteristics calculation formula is as follows: ； in Indicates that the y-th marketing plan is equivalent to the marketing characteristics of the i-th labeled entity, represents the set of historically sold models of the y-th marketing plan, and d represents The total number of models, express The car text features of the p-th car model in, represents the car model parameter features of the i-th labeled entity, represents the similarity function; The similarity calculation formula is as follows: ； in, represents the similarity value between the car parameter features of the ith labeled entity and the marketing features of the yth marketing plan relative to the ith labeled entity, represents the similarity function; The marketing plan recommendation module removes duplicate marketing plans and obtains a final set of recommended marketing plans; The cost accounting module calculates the total cost of all plans based on the costs involved in the marketing plan; The marketing program scoring module is used to record the final effect score of the marketing program used by the current model and to calculate the score, and the result of the statistics is the average score.

10. A storage medium storing non-transitory computer-readable instructions, characterized in that: When the non-transitory computer-readable instructions are executed by a computer, the steps in the method for monitoring automobile marketing data as described in any one of claims 1 to 8 can be executed.

Images