CN114398486B

CN114398486B - Method and device for intelligently customizing customer acquisition publicity

Info

Publication number: CN114398486B
Application number: CN202210013837.4A
Authority: CN
Inventors: 徐勇; 胡鑫平; 陈钰; 刘作来
Original assignee: Beijing Borui Tongyun Technology Co ltd
Current assignee: Beijing Borui Tongyun Technology Co ltd
Priority date: 2022-01-06
Filing date: 2022-01-06
Publication date: 2022-08-26
Anticipated expiration: 2042-01-06
Also published as: CN114398486A

Abstract

The embodiment of the invention relates to a method and a device for intelligently customizing a passenger obtaining publicity, wherein the method comprises the following steps: acquiring first passenger acquisition publicity language data and corresponding first activity scene data; converting the text characteristic vector to generate a first propaganda text characteristic vector; constructing a first scene feature vector according to the first product type data and the first activity type data; carrying out propaganda effect value prediction on the first propaganda text characteristic vector and the first scene characteristic vector by using a propaganda effect identification model to generate first effect value data; recording the first passenger advertisement data as qualified advertisement data when the first effect value data exceeds the evaluation effect value threshold; recording the first passenger acquisition publicity language data as unqualified publicity data when the evaluation effect value threshold value is not exceeded; and when the first passenger obtaining propaganda data is evaluated to be qualified propaganda data, outputting the first passenger obtaining propaganda data as the customized passenger obtaining propaganda. By the method and the device, the efficiency and the quality of publicity customization can be improved.

Description

Method and device for intelligently customizing customer acquisition publicity

Technical Field

The invention relates to the field of natural language processing, in particular to a method and a device for intelligently customizing a customer-obtaining publicity.

Background

Publicity is seen everywhere in human commercial marketing activities, core concepts closely related to a specific marketing purpose can be accurately expressed by concise text contents based on the purpose when the customer publicity is customized, the marketing effect for playing the advertisement is usually processed by adopting a large-range releasing mode when the customer publicity is released, and large-range releasing is bound to generate larger publicity cost. In order to ensure that the invested propaganda cost is in direct proportion to the brought reverberation effect, the customized customer acquisition propaganda speech quality needs to be reasonably controlled. At present, the quality control method of each company for the customer publicity words related to the commercial marketing activities and products of each company is mainly based on the manual experience of related markets and advertisers, and a stable customized quality assessment means which is separated from manual intervention cannot be provided.

Disclosure of Invention

The invention aims to provide a method, a device, electronic equipment and a computer readable storage medium for intelligently customizing a customer publicity Language, which aim to overcome the defects of the prior art, perform feature recognition on a publicity Language to be evaluated by using a Natural Language Processing (NLP) technology to obtain a corresponding text feature vector, perform feature recognition on commercial activity scene information corresponding to the publicity Language to obtain a corresponding scene feature vector, and perform effect value prediction on the text feature vector and the scene feature vector by using a publicity effect recognition model, so that qualified or unqualified effect evaluation can be quickly given to each publicity Language to be selected on the premise of not depending on any artificial experience, and the publicity Language evaluated as qualified is output as a customization result. By the method and the device, the degree of dependence on manual experience in the quality control process of the customized passenger publicity phrases can be reduced, and the purpose of improving the customization quality of the publicity phrases can be achieved by improving the quality control efficiency of the publicity phrases.

In order to achieve the above object, a first aspect of the embodiments of the present invention provides a method for intelligently customizing a guest advertisement, where the method includes:

acquiring first passenger acquisition publicity language data and corresponding first activity scene data; the first activity scenario data comprises first product type data and first activity type data;

performing text feature vector conversion processing on the first passenger propaganda data to generate a corresponding first propaganda text feature vector;

constructing a corresponding first scene feature vector according to the first product type data and the first activity type data;

carrying out propaganda effect value prediction on the first propaganda text characteristic vector and the first scene characteristic vector by using a well-trained propaganda effect identification model to generate corresponding first effect value data;

when the first effect value data exceeds a preset evaluation effect value threshold value, recording the corresponding first passenger obtaining propaganda data as qualified propaganda data; when the first effect value data does not exceed the evaluation effect value threshold value, marking the corresponding first passenger obtaining propaganda data as unqualified propaganda data;

and when the first passenger obtaining propaganda data is evaluated to be qualified propaganda data, outputting the first passenger obtaining propaganda data as the customized passenger obtaining propaganda.

Preferably, the performing text feature vector conversion processing on the first passenger advertisement data to generate a corresponding first advertisement text feature vector specifically includes:

based on a preset sentence break symbol, carrying out sentence break processing on the first passenger publicity language data to generate a corresponding first sentence data sequence; the first sentence data sequence includes a plurality of first sentence data;

based on a preset intelligent word segmentation model, performing word segmentation and part-of-speech tagging on each first sentence data to generate a corresponding first word segmentation data group sequence, and forming a first word segmentation data group sequence set by all the obtained first word segmentation data group sequences; the first sequence of partial word data sets comprises a plurality of first partial word data sets; the first word segmentation data group comprises first word segmentation text data and first word segmentation data;

performing text grammar evaluation processing on the first word segmentation data group sequence set to generate corresponding first grammar evaluation data;

performing text emotion assessment processing on the first word segmentation data group sequence set based on a preset text emotion analysis model to generate corresponding first emotion assessment data; the first sentiment assessment data comprises at least a positive sentiment state value and a negative sentiment state value;

performing text style confirmation processing on the first activity scene data to generate corresponding first style data;

performing text style matching degree calculation processing on the first word segmentation data group sequence set according to the first style data to generate corresponding first style matching degree data;

performing scene keyword weight calculation processing on the first participle data set sequence set according to the first activity scene data to generate corresponding first scene keyword weight data;

and constructing the first publicity text feature vector according to the first grammar evaluation data, the first emotion evaluation data, the first style matching degree data and the first scene keyword weight data.

Further, the performing text grammar evaluation processing on the first word segmentation data group sequence set to generate corresponding first grammar evaluation data specifically includes:

based on a preset text grammar template, carrying out grammar evaluation on each first participle data group sequence of the first participle data group sequence set to generate corresponding first sentence grammar evaluation data; the first sentence grammar evaluation data comprises pass and fail;

counting the number of the first participle data set sequences in the first participle data set sequence set to generate a first sentence total number; counting the number of the first sentence grammar evaluation data which are qualified to generate a first qualified sentence total number; generating a first qualified rate according to the ratio of the total number of the first qualified sentences to the total number of the first sentences;

according to a preset corresponding relation between the qualification rate and the grammar level, obtaining grammar level state value information corresponding to the first qualification rate, and generating first sentence grammar evaluation data; the first sentence grammar evaluation data includes a poor level status value, a medium level status value, a good level status value, and a good level status value.

Further, the performing text style confirmation processing on the first activity scene data to generate corresponding first style data specifically includes:

extracting a first product text style range field of a first product style data record of which a first product type field is matched with the first product type data of the first activity scene data from a preset first product style database as a corresponding first text style range; the first product style database comprises a plurality of the first product style data records; the first product style data record comprises the first product type field and the first product text style range field; the first product text style range field comprises a plurality of text style types;

extracting a first activity text style range field of a first activity style data record of which a first activity type field is matched with the first activity type data of the first activity scene data from a preset first activity style database to serve as a corresponding second text style range; said first activity style database comprises a plurality of said first activity style data records; the first activity style data record comprises the first activity type field and the first activity text style range field; the first active text style range field comprises a plurality of text style types;

generating the first style data by collecting the text style types of the first text style range and the second text style range; the first style data includes one or more text style types.

Further, the performing text style matching degree calculation processing on the first segmentation data group sequence set according to the first style data to generate corresponding first style matching degree data specifically includes:

based on a preset text style classification model, performing text style classification processing on the first word segmentation data group sequence set to obtain a plurality of first classification probabilities corresponding to the first classification data; selecting the first classification data with the maximum first classification probability as first publicity type data;

if the first style data has a text style type matched with the first publicity type data, setting the first style matching degree data as a matching state value; and if the text style type matched with the first publicity type data does not exist in the first style data, setting the first style matching degree data as a mismatching state value.

Further, the performing scene keyword weight calculation processing on the first segmentation data group sequence set according to the first activity scene data to generate corresponding first scene keyword weight data specifically includes:

extracting a first product keyword range field of a first product keyword data record of which a second product type field is matched with the first product type data of the first activity scene data from a preset first product keyword database to serve as a corresponding first keyword range; the first product keyword database comprises a plurality of the first product keyword data records; the first product keyword data record comprises the second product type field and the first product keyword range field; the first product keyword range field comprises a plurality of keywords and keyword weights corresponding to the keywords; the first keyword range includes a plurality of first keyword data groups; the first keyword data group comprises first keywords and first keyword weights;

extracting a first activity keyword range field of a first activity keyword data record of which a second activity type field is matched with the first activity type data of the first activity scene data from a preset first activity keyword database to serve as a corresponding second keyword range; the first active keyword database comprises a plurality of the first active keyword data records; the first campaign keyword data record comprises the second campaign type field and the first campaign keyword range field; the first activity keyword range field comprises a plurality of keywords and corresponding keyword weights; the second keyword range includes a plurality of second keyword data sets; the second keyword data group comprises second keywords and second keyword weights;

extracting the first keyword data group of which the first keyword does not appear in the second keyword range as a third keyword data group; extracting the second keyword data group of which the second keyword does not appear in the first keyword range to serve as the third keyword data group; the third keyword data group comprises third keywords and third keyword weights;

performing keyword data combination and processing on first and second keyword data groups corresponding to the first and second keywords with the same content, extracting any keyword as a corresponding third keyword, calculating and generating a corresponding third keyword weight according to the first and second keyword weights, and forming a corresponding third keyword data group by the obtained third keyword and the third keyword weight; the third keyword weight value (a + first keyword weight value + b + second keyword weight value)/(a + b);

forming a third keyword range by obtaining all the third keyword data groups;

polling each third keyword data group in the third keyword range, recording the currently polled third keyword data group as a current keyword data group, recording the third keywords in the current keyword data group as current keywords, and recording the third keyword weight of the current keyword data group as a current keyword weight; counting the number of the first word segmentation text data matched with the current keywords in the first word segmentation data group sequence set to generate the number of the current keywords; generating corresponding first keyword weight data according to the product of the current keyword weight and the number of the current keywords;

and performing sum calculation on all the obtained first keyword weight data to generate the first scene keyword weight data.

Preferably, before the using the trained mature promotional effect recognition model, the method further comprises training the promotional effect recognition model, specifically:

acquiring a plurality of groups of first historical data groups; the first historical data set comprises first historical publicity language data, first historical activity scene data and first historical publicity effect data; the first historical activity scenario data comprises second product type data and second activity type data; the first historical publicity effect data comprises a first activity page click total amount, a first activity participant total amount and a first activity product sales total amount;

performing text feature vector conversion processing on the first historical propaganda data of each first historical data group to generate corresponding first historical propaganda text feature vectors; constructing a corresponding first historical scene feature vector according to the second product type data and the second activity type data of each first historical data group; the click total amount of the first activity page, the total amount of the first activity participants and the total amount of the first activity product sales of each first historical data group are brought into a preset propaganda effect value calculation formula for calculation, and corresponding first historical effect value data are generated; each group of the first historical propaganda text characteristic vectors, the first historical scene characteristic vectors and the first historical effect value data form a corresponding first training data group;

inputting the first historical propaganda text feature vector and the first historical scene feature vector of the first training data set into the propaganda effect recognition model for training to generate corresponding first training effect value data;

performing error calculation on the first training effect value data and the first historical effect value data of the corresponding first training data set based on a preset model error calculation function to generate first effect value error data;

when the first effect value error data is not in a preset reasonable effect value error range, modulating the neural network parameters and/or the neural network structure of the propaganda effect recognition model according to the first effect value error data, and continuing to train the propaganda effect recognition model by using the next group of first training data set after modulation;

and when the first effect value error data is within the reasonable effect value error range, stopping training the propaganda effect recognition model, and setting the training degree of the propaganda effect recognition model as training maturity.

A second aspect of the embodiments of the present invention provides an apparatus for intelligently customizing a guest-obtaining advertisement, including: the device comprises an acquisition module, a feature vector preparation module, a propaganda effect prediction module, a propaganda data qualification judgment module and a propaganda language customization output module;

the acquisition module is used for acquiring first passenger acquisition publicity language data and corresponding first activity scene data; the first activity scenario data comprises first product type data and first activity type data;

the characteristic vector preparation module is used for performing text characteristic vector conversion processing on the first passenger propaganda data to generate a corresponding first propaganda text characteristic vector; constructing a corresponding first scene feature vector according to the first product type data and the first activity type data;

the propaganda effect prediction module is used for predicting propaganda effect values of the first propaganda text characteristic vector and the first scene characteristic vector by using a well-trained propaganda effect recognition model to generate corresponding first effect value data;

the propaganda data qualification judging module is used for recording the corresponding first passenger propaganda data as qualified propaganda data when the first effect value data exceeds a preset evaluation effect value threshold; when the first effect value data does not exceed the evaluation effect value threshold value, marking the corresponding first passenger obtaining propaganda data as unqualified propaganda data;

and the propaganda customization output module is used for outputting the first passenger propaganda data as the customized passenger propaganda data when the first passenger propaganda data is evaluated to be qualified propaganda data.

A third aspect of an embodiment of the present invention provides an electronic device, including: a memory, a processor, and a transceiver;

the processor is configured to be coupled to the memory, read and execute instructions in the memory, so as to implement the method steps of the first aspect;

the transceiver is coupled to the processor, and the processor controls the transceiver to transmit and receive messages.

A fourth aspect of embodiments of the present invention provides a computer-readable storage medium storing computer instructions that, when executed by a computer, cause the computer to perform the method of the first aspect.

The embodiment of the invention provides a method, a device, electronic equipment and a computer readable storage medium for intelligently customizing a customer-obtaining publicity, which refer to an NLP technology to perform feature recognition on a publicity to be evaluated to obtain a corresponding text feature vector, perform feature recognition on commercial activity scene information corresponding to the publicity to obtain a corresponding scene feature vector, and then use a publicity effect recognition model to perform effect value prediction on the text feature vector and the scene feature vector, so that qualified or unqualified effect evaluation can be quickly given to each publicity to be selected on the premise of not depending on any artificial experience. By the method and the device, the degree of dependence on manual experience in the quality control process of the customized passenger publicity is reduced, and the efficiency and the quality of publicity customization are improved.

Drawings

Fig. 1 is a schematic diagram illustrating a method for intelligently customizing a target advertisement according to an embodiment of the present invention;

FIG. 2 is a block diagram of an apparatus for intelligently customizing a target advertisement according to a second embodiment of the present invention;

fig. 3 is a schematic structural diagram of an electronic device according to a third embodiment of the present invention.

Detailed Description

In order to make the objects, technical solutions and advantages of the present invention clearer, the present invention will be described in further detail with reference to the accompanying drawings, and it is apparent that the described embodiments are only a part of the embodiments of the present invention, not all of the embodiments. All other embodiments, which can be obtained by a person skilled in the art without making any creative effort based on the embodiments in the present invention, belong to the protection scope of the present invention.

Fig. 1 is a schematic view of a method for intelligently customizing a guest-obtaining advertisement according to an embodiment of the present invention, and as shown in fig. 1, the method mainly includes the following steps:

step 1, acquiring first passenger propaganda data and corresponding first activity scene data;

wherein the first activity scenario data includes first product type data and first activity type data.

Here, the first subject publicity language data is text data of the subject publicity language which is currently used for evaluation; the first product type data is used for identifying a specific product type, the insurance product may include a property insurance type, a vehicle insurance type, a life insurance type and the like by taking the insurance product as an example, different numbers are used for identifying different product types under the conventional condition, and the insurance product indicates that the product type is the property insurance type when the first product type data is 1, the product type is the vehicle insurance type when the first product type data is 2 and the product type is the life insurance type when the first product type data is 3 by taking the insurance product as an example; the first activity type data is used to identify specific activity types, and the promotional activity may include daily promotion types, holiday promotion types, end-of-year promotion types, etc. different numbers are conventionally used to identify different activity types, and the promotional activity may indicate that the activity type is a daily promotion type when the first activity type data is 1, a holiday promotion type when the first activity type data is 2, and an end-of-year promotion type when the first activity type data is 3.

Step 2, performing text feature vector conversion processing on the first passenger propaganda language data to generate a corresponding first propaganda text feature vector;

the method specifically comprises the following steps: step 21, based on a preset sentence break symbol, carrying out sentence break processing on the first passenger propaganda language data to generate a corresponding first sentence data sequence;

wherein the first sentence data sequence includes a plurality of first sentence data; the punctuation marks comprise period marks, semicolon marks, ellipsis marks, exclamation marks and question marks; here, the punctuation can be increased or decreased according to the practical application condition;

here, the current step is to perform data preprocessing operations related to special symbol filtering and text sentence breaking on the first guest publicity language data;

the method specifically comprises the following steps: step 211, based on preset special characters, performing special character filtering processing on the first passenger publicity language data to generate first filtered publicity language data;

the special characters comprise space symbols, carriage return symbols, line feed symbols, meter-making position symbols and other special symbols related to a text editing format;

here, the conditions of line segmentation, space and the like do not occur in the first filtered publicity language data obtained by special character filtering processing;

step 212, sequentially identifying all punctuation marks of the first filtered poster data, and marking the current character as a punctuation character if the current punctuation mark is matched with any one of the punctuation marks;

for example, the first filtered hyphen data is "I see an apple! That apple was struck on the ground by a kitten … … "with the punctuation mark"! "matches the exclamation mark symbol in the preset break and the punctuation mark" … … "matches the ellipsis mark in the preset break, then the punctuation mark"! "and" … … "will be marked as sentence break characters;

step 213, in the first filtering publicity language data, using the character string data from the initial character position to the previous character position of the first sentence-punctuating character as the first sentence data; sequentially extracting character string data between two adjacent sentence-breaking characters from the first sentence-breaking character as corresponding first sentence data;

for example, the first filtered hyphen data is "I see an apple! That apple was hit on the ground by a kitten … … ", punctuation mark! "and" … … "are punctuation characters, then 2 first sentence data can be obtained, which are the first sentence data 1" i see an apple "and the first sentence data 2" that apple was hit by a kitten on the ground ";

step 214, sequencing all the obtained first sentence data according to the sequence to generate a first sentence data sequence;

step 22, based on a preset intelligent word segmentation model, performing word segmentation and part-of-speech tagging on each first sentence data to generate a corresponding first word segmentation data group sequence, and forming a first word segmentation data group sequence set by all the obtained first word segmentation data group sequences;

wherein the first sequence of participle data sets comprises a plurality of first participle data sets; the first word segmentation data group comprises first word segmentation text data and first word segmentation data;

here, the current step is data preprocessing for performing single character/word segmentation on each short sentence text in the first guest publicity language data, that is, the first sentence data;

the preset intelligent word segmentation model comprises a word segmentation module and a part-of-speech tagging module; the word segmentation module is an artificial intelligent word segmentation Model constructed based on an NLP (non line segment) word segmentation algorithm, and commonly comprises a word segmentation Model based on a jieba (jieba) algorithm, a word segmentation Model based on a Hidden Markov Model (HMM) algorithm, a word segmentation Model based on a Conditional Random Field (CRF) algorithm and a word segmentation Model based on a Long Short-Term Memory network (LSTM) algorithm; the part-of-speech tagging module is a part-of-speech recognition and tagging processing module based on a preset dictionary, and has the functions of specifically acquiring part-of-speech information of each first-word segmentation text data by querying the preset dictionary, setting the first-word segmentation data corresponding to each first-word segmentation text data according to the part-of-speech information acquired by querying so as to acquire a plurality of first-word segmentation data groups, and sequencing the plurality of first-word segmentation data groups according to the sequence positions of the first-word segmentation text data in the first text data so as to acquire a first-word segmentation data group sequence;

step 23, performing text grammar evaluation processing on the first word segmentation data group sequence set to generate corresponding first grammar evaluation data;

the first sentence grammar evaluation data comprises a qualified sentence and an unqualified sentence;

here, the current step is actually checking the grammar of each short sentence text in the first guest publicity language data, and obtaining the full-text grammar checking result of the first guest publicity language data, that is, the first grammar evaluation data, based on the grammar checking results of all the short sentence texts;

the method specifically comprises the following steps: 231, performing grammar evaluation on each first participle data group sequence of the first participle data group sequence set based on a preset text grammar template to generate corresponding first sentence grammar evaluation data;

here, each first segmentation data group sequence corresponds to a short sentence text; the text grammar template includes one or more sub-grammar templates, each sub-grammar template giving a set of grammar structures; each grammar structure gives out a plurality of grammar bodies with sequence relations, and each grammar body is configured with part-of-speech limited relations; when grammar evaluation is carried out, a plurality of first sub-word part-of-speech data in the first sub-word data group sequence are sequentially arranged to generate a first part-of-speech sequence, then each sub-grammar template is used for carrying out grammar qualification judgment on the first part-of-speech sequence, and if the grammar qualification judgment of all the sub-grammar templates is qualified, the first sentence grammar evaluation data is qualified;

for example, if the first word data group sequence corresponding to the first sentence data "i eat" is { the first word data group [ "i", pronouns ], the first word data group [ "eat", verbs ], the first word data group [ "rice", nouns ] }, then the corresponding first part-of-speech sequence should be { pronouns, verbs, nouns };

the sub-grammar template provides a subject-predicate grammar structure of a subject grammar body + a predicate grammar body + an object grammar body, the subject grammar body is configured with a part-of-speech restricted relation which can be pronouns and nouns, the predicate grammar body is configured with a part-of-speech restricted relation which can be verbs, and the object grammar body is configured with a part-of-speech restricted relation which can be pronouns and nouns; then, the part-of-speech sequence satisfying the sub-grammar template may have 4 kinds of { pronouns, verbs, nouns }, { nouns, verbs, nouns }, { pronouns, verbs, pronouns }, { nouns, verbs, pronouns };

the grammar qualification judgment is carried out on the first part-of-speech sequence { pronouns, verbs and nouns } on the basis of the sub-grammar template, and whether the first part-of-speech sequence is one of the 4 part-of-speech sequences is actually judged, in the example, the first part-of-speech sequence { pronouns, verbs and nouns } is the 1 st of the 4 part-of-speech sequences, so that the obtained judgment result, namely the first sentence grammar evaluation data is qualified;

step 232, counting the number of the first branch word data group sequences in the first branch word data group sequence set to generate a first sentence total number; counting the number of the first sentence grammar evaluation data which are qualified to generate the total number of the first qualified sentences; generating a first qualified rate according to the ratio of the total number of the first qualified sentences to the total number of the first sentences;

for example, if there are 4 first participle data set sequences in the first participle data set sequence set, the total number of the first sentences is 4; specifically, if the number of the qualified first sentence grammar evaluation data is 3, the total number of the first qualified sentences is 3; then the first pass rate is 75%;

step 233, according to the preset corresponding relationship between the qualification rate and the grammar level, obtaining grammar level state value information corresponding to the first qualification rate, and generating first sentence grammar evaluation data;

wherein the first sentence grammar evaluation data includes a poor level status value, a medium level status value, a good level status value, and a good level status value;

here, the preset corresponding relationship between the qualification rate and the grammar level is specifically realized through a corresponding relationship table, the corresponding relationship table comprises a plurality of corresponding relationship records, each corresponding relationship record comprises a qualification rate range and grammar level state value information, and the grammar level state value information can be a poor level state value, a medium level state value, a good level state value or an excellent level state value; when grammar level state value information corresponding to the first qualified rate is obtained, recording a corresponding relation corresponding to a qualified rate range which is satisfied by the first qualified rate in the corresponding relation table as a matching relation record, and extracting the grammar level state value information recorded by the matching relation as first sentence grammar evaluation data;

for example, the correspondence table has 4 correspondence records, which are respectively: correspondence record 1 (the pass rate range is < 60%, the syntax level state value information is a poor level state value), correspondence record 2 (the pass rate range is 61% -80%, the syntax level state value information is a medium level state value), correspondence record 3 (the pass rate range is 81% -90%, the syntax level state value information is a good level state value), correspondence record 4 (the pass rate range is 91% -100%, the syntax level state value information is a good level state value);

then, when the first pass rate is 75%, the obtained first sentence grammar evaluation data is a medium-level state value;

step 24, performing text emotion assessment processing on the first word segmentation data group sequence set based on a preset text emotion analysis model to generate corresponding first emotion assessment data;

wherein the first sentiment assessment data includes at least a positive sentiment state value and a negative sentiment state value;

here, the current step is actually analyzing the emotion component of the first passenger poster data; the text emotion Analysis model is actually a text emotion Analysis (Sentiment Analysis) model based on an NLP technology, and the function of the model is to calculate emotion/emotion weight of a text based on known emotion/emotion keywords and/or emotion/emotion phrases and classify the text according to the obtained emotion/emotion weight; the classification algorithm commonly used by the model comprises a central vector classification algorithm, a KNN (K-Nearest-Neighbor) class algorithm, a Bayesian classification algorithm, a support vector machine classification algorithm, a conditional random field classification algorithm and a maximum entropy classification algorithm;

step 25, performing text style confirmation processing on the first activity scene data to generate corresponding first style data;

here, the current step is actually determining the text style of the required promotional text according to the campaign type and the product information of the current commercial marketing campaign, i.e., the first campaign scenario data;

the method specifically comprises the following steps: step 251, extracting a first product text style range field of a first product style data record of which the first product type field is matched with first product type data of the first activity scene data from a preset first product style database as a corresponding first text style range;

wherein the first product style database comprises a plurality of first product style data records; the first product style data record comprises a first product type field and a first product text style range field; the first product text style range field includes a plurality of text style types;

here, the embodiment of the present invention will summarize the best matching one or more text styles for different product types based on long-term data deposition, and thus establish the first product style database;

step 252, extracting a first activity text style range field of the first activity style data record, of which the first activity type field is matched with the first activity type data of the first activity scene data, from a preset first activity style database, and taking the first activity text style range field as a corresponding second text style range;

wherein the first activity style database comprises a plurality of first activity style data records; the first activity style data record comprises a first activity type field and a first activity text style range field; the first active text style range field includes a plurality of text style types;

here, the embodiment of the present invention may summarize the best matching one or more text styles for different marketing campaign types based on long-term data deposition, and thus establish the first campaign style database;

step 253, a text style type collection of the first text style range and the second text style range is carried out to generate first style data;

wherein the first style data comprises one or more text style types;

for example, if the first text style range is (concise, digitized), and the second text style range is (concise, plain), the first style data is (concise, digitized, plain), that is, for the commercial marketing campaign, the text style of the required promotional literature should be characterized by being concise, digitized, plain, and the like;

step 26, performing text style matching degree calculation processing on the first word segmentation data group sequence set according to the first style data to generate corresponding first style matching degree data;

here, the matching relationship between the text style of the first customer publicity language data and the text style required by the commercial marketing campaign is actually analyzed in the current step;

the method specifically comprises the following steps: 261, based on a preset text style classification model, performing text style classification processing on the first word segmentation data group sequence set to obtain a plurality of first classification probabilities corresponding to the first classification data; selecting first classification data with the first classification probability as the maximum value as first publicity type data;

here, the text style classification model may specifically be a text classification model based on a Convolutional Neural Network (CNN), a text classification model based on a Recurrent Neural Network (RNN), and a text classification model based on a Long Short-Term-Memory artificial Neural Network (LSTM) + Attention mechanism; the embodiment of the invention uses a text style classification model to calculate the text style weight of a first word data group sequence set, then carries out classification probability conversion according to the text style weight to obtain probability values corresponding to a plurality of types, namely first classification probabilities corresponding to a plurality of first classification data, and finally takes the type corresponding to the maximum probability as a final analysis type to be output, namely first publicity type data;

for example, the text style classification model may output classification probabilities of three text styles, namely a concise style probability, a graceful style probability, and a luxury style probability; specific values of the currently output 3 probabilities are a concise style probability of 85%, a graceful style probability of 15%, and a luxury style probability of 5%, so that the first poster type data should be a text style corresponding to the maximum probability (the concise style probability of 85%), namely concise;

step 262, if the first style data has a text style type matched with the first publicity type data, setting the first style matching degree data as a matching state value; if the text style type matched with the first publicity type data does not exist in the first style data, setting the first style matching degree data as a mismatching state value;

for example, the first style data is (concise, digitized, plain) and the first poster type data is concise, then the first style data has a text style type matching with the first poster type data, and the corresponding first style matching degree data should be set as a matching state value;

step 27, performing scene keyword weight calculation processing on the first participle data set sequence set according to the first activity scene data to generate corresponding first scene keyword weight data;

in the current step, the fitting relationship between the text content of the first customer advertisement language data and the commercial marketing campaign is actually analyzed, the higher the first scene keyword weight data is, the higher the fitting degree is, otherwise, the smaller the first scene keyword weight data is, the smaller the fitting degree is;

the method specifically comprises the following steps: step 271, extracting a first product keyword range field of a first product keyword data record, of which the second product type field matches with the first product type data of the first activity scene data, from a preset first product keyword database as a corresponding first keyword range;

wherein the first product keyword database comprises a plurality of first product keyword data records; the first product keyword data record comprises a second product type field and a first product keyword range field; the first product keyword range field comprises a plurality of keywords and keyword weights corresponding to the keywords; the first keyword range includes a plurality of first keyword data groups; the first keyword data group comprises first keywords and first keyword weights;

here, the embodiment of the present invention summarizes a series of associated keywords and corresponding keyword weights for different product types based on long-term data deposition, and thereby establishes the first product keyword database; each keyword weight reflects the relevance between the corresponding keyword and the corresponding product type, the higher the keyword weight is, the higher the relevance is, otherwise, the smaller the keyword weight is, the smaller the relevance is;

step 272, extracting a first activity keyword range field of a first activity keyword data record, of which the second activity type field is matched with the first activity type data of the first activity scene data, from a preset first activity keyword database as a corresponding second keyword range;

wherein the first campaign keyword database comprises a plurality of first campaign keyword data records; the first activity keyword data record comprises a second activity type field and a first activity keyword range field; the first activity keyword range field comprises a plurality of keywords and corresponding keyword weights; the second keyword range includes a plurality of second keyword data sets; the second keyword data group comprises second keywords and second keyword weights;

here, the embodiment of the present invention summarizes a series of associated keywords and corresponding keyword weights for different activity types based on long-term data deposition, and thereby establishes the first activity keyword database; each keyword weight reflects the relevance between the corresponding keyword and the corresponding activity type, the higher the keyword weight is, the higher the relevance is, otherwise, the smaller the keyword weight is, the smaller the relevance is;

step 273, extracting a first keyword data group of which the first keyword does not appear in the second keyword range as a third keyword data group; extracting a second keyword data group of which the second keyword does not appear in the first keyword range to serve as a third keyword data group;

the third keyword data group comprises third keywords and third keyword weights;

step 274, performing keyword data combination and processing on first and second keyword data sets corresponding to first and second keywords with the same content, extracting any keyword from the first and second keyword data sets as a corresponding third keyword, calculating and generating a corresponding third keyword weight according to the first and second keyword weights, and forming a corresponding third keyword data set by the obtained third keyword and the third keyword weight;

wherein, the third keyword weight value (a + first keyword weight value + b + second keyword weight value)/(a + b);

275, forming a third keyword range by obtaining all the third keyword data groups;

here, the step 273-; the finally obtained third keyword range is actually the keyword corpus with the highest association degree with the activity type and the product type of the commercial marketing activity;

step 276, polling each third keyword data group in the third keyword range, recording the currently polled third keyword data group as a current keyword data group, recording the third keywords of the current keyword data group as current keywords, and recording the third keyword weight of the current keyword data group as the current keyword weight; counting the number of first word segmentation text data in the first word segmentation data group sequence set and the number of first word segmentation data groups matched with the current keywords to generate the number of the current keywords; generating corresponding first keyword weight data according to the product of the current keyword weight and the current keyword quantity;

here, the first keyword weight data reflects the engagement degree of each short sentence text content in the first passenger promotion data with the commercial marketing campaign;

277, performing sum calculation on all the obtained first keyword weight data to generate first scene keyword weight data;

here, the first scenario keyword weight data reflects the fitness of the full text content of the first guest publicity language data and the commercial marketing campaign;

and step 28, constructing a first publicity text feature vector according to the first grammar evaluation data, the first emotion evaluation data, the first style matching degree data and the first scene keyword weight data.

The first publicity text feature vector comprises feature data reflecting the text style requirement of the commercial marketing activity, namely first style data; the feature data reflecting the grammar quality of the publicity text, namely the first grammar evaluation data, the feature data reflecting the emotional tendency of the publicity text, namely the first emotion evaluation data, the feature data reflecting the matching between the style of the publicity text and the scene, namely the first style matching degree data, and the feature data reflecting the matching between the content of the publicity text and the scene, namely the first scene keyword weight data are also included.

And 3, constructing a corresponding first scene feature vector according to the first product type data and the first activity type data.

Here, the first scene feature vector contains only two feature data, the first product type data and the first activity type data.

For example, the vector structure of the first scene feature vector is [ product type, activity type ], the first product type data is 2, the first activity type data is 3, and then the first scene feature vector is specifically [2, 3 ].

And 4, carrying out propaganda effect value prediction on the first propaganda text characteristic vector and the first scene characteristic vector based on the well-trained propaganda effect recognition model, and generating corresponding first effect value data.

Here, the advertisement effect recognition model according to the embodiment of the present invention is used to predict an advertisement effect of the current commercial marketing campaign that is advertised using the first guest advertisement data, and the higher the first effect value data is, the better the effect is, whereas the lower the first effect value data is, the worse the effect is.

It should be noted that, training the promotional effectiveness recognition model is required before using the promotional effectiveness recognition model, and specifically, training the promotional effectiveness recognition model includes:

step A1, acquiring a plurality of groups of first historical data groups;

the first historical data group comprises first historical propaganda language data, first historical activity scene data and first historical propaganda effect data; the first historical activity scenario data includes second product type data and second activity type data; the first historical promotional effect data comprises a first campaign page click total, a first campaign participant total and a first campaign product sale total;

step A2, performing text feature vector conversion processing on the first historical propaganda data of each group of first historical data groups to generate corresponding first historical propaganda text feature vectors; constructing a corresponding first historical scene feature vector according to the second product type data and the second activity type data of each first historical data group; the click total amount of the first activity page, the total amount of the first activity participants and the total amount of the first activity product sales of each first historical data group are brought into a preset propaganda effect value calculation formula for calculation, and corresponding first historical effect value data are generated; each group of first historical propaganda text characteristic vectors, first historical scene characteristic vectors and first historical effect value data form a corresponding first training data group;

step A3, inputting a first historical propaganda text characteristic vector and a first historical scene characteristic vector of a first training data set into a propaganda effect recognition model for training, and generating corresponding first training effect value data;

step A4, performing error calculation on the first training effect value data and the first historical effect value data of the corresponding first training data set based on a preset model error calculation function to generate first effect value error data;

step A5, when the first effect value error data is not in the preset reasonable effect value error range, modulating the neural network parameters and/or the neural network structure of the propaganda effect recognition model according to the first effect value error data, and continuing to train the propaganda effect recognition model by using the next group of first training data group after modulation;

and step A6, stopping training the propaganda effect recognition model when the first effect value error data is in the reasonable effect value error range, and setting the training degree of the propaganda effect recognition model as training maturity.

Step 5, when the first effect value data exceeds a preset evaluation effect value threshold value, recording the corresponding first passenger obtaining propaganda data as qualified propaganda data; and when the first effect value data does not exceed the evaluation effect value threshold value, recording the corresponding first passenger advertisement data as unqualified advertisement data.

Here, the evaluation effect value threshold is a preset system threshold, and if the evaluation effect value threshold exceeds the preset system threshold, it indicates that the promotion prediction effect of the first guest advertising data meets the guest advertising expectation of the current commercial marketing campaign, and if the evaluation effect value threshold does not exceed the preset system threshold, it indicates that the promotion prediction effect of the first guest advertising data does not meet the guest advertising expectation of the current commercial marketing campaign.

And 6, when the first passenger obtaining propaganda data is evaluated to be qualified propaganda data, outputting the first passenger obtaining propaganda data as the customized passenger obtaining propaganda.

Here, the passenger obtaining effect qualification evaluation result of the first passenger obtaining publicity language data to be evaluated at this time, namely qualified publicity data or unqualified publicity data, is obtained through the above steps 1-5; and if the evaluation result is qualified propaganda data, outputting the first passenger propaganda data to be evaluated at the time as a customized passenger propaganda text.

It should be noted that, each time the guest obtaining propaganda material is customized, a plurality of pieces of guest obtaining propaganda material data to be evaluated may be generated in advance, and then each guest obtaining propaganda material to be evaluated may be analyzed in sequence through the above steps 1 to 6, and the first guest obtaining propaganda material data evaluated as qualified propaganda material is taken into the range of the customizable guest obtaining propaganda material, and then the first guest obtaining propaganda material data with the best first effect value data is selected as the finally selected customized guest obtaining propaganda material data.

Fig. 2 is a block diagram of a device for intelligently customizing a guest-obtaining advertisement according to a second embodiment of the present invention, where the device may be a terminal device or a server for implementing the method according to the second embodiment of the present invention, or may be a device connected to the terminal device or the server for implementing the method according to the second embodiment of the present invention, and for example, the device may be a device or a chip system of the terminal device or the server. As shown in fig. 2, the apparatus includes: the system comprises an acquisition module 201, a feature vector preparation module 202, a promotion effect prediction module 203, a promotion data qualification judgment module 204 and a promotion customization output module 205.

The obtaining module 201 is configured to obtain first guest obtaining publicity data and corresponding first activity scene data; the first activity scenario data includes first product type data and first activity type data.

The feature vector preparation module 202 is configured to perform text feature vector conversion processing on the first guest-obtaining publicity language data to generate a corresponding first publicity text feature vector; and constructing a corresponding first scene feature vector according to the first product type data and the first activity type data.

The promotion effect prediction module 203 is configured to perform promotion effect value prediction on the first promotion text feature vector and the first scene feature vector using a well-trained promotion effect recognition model, and generate corresponding first effect value data.

The promotion data eligibility determination module 204 is configured to mark the corresponding first guest-obtaining promotion data as eligible promotion data when the first effect value data exceeds a preset evaluation effect value threshold; and when the first effect value data does not exceed the evaluation effect value threshold value, recording the corresponding first passenger advertisement data as unqualified advertisement data.

The advertisement customization output module 205 is configured to output the first advertisement data as the customized advertisement when the first advertisement data is evaluated as qualified advertisement data.

The device for intelligently customizing the object obtaining publicity language provided by the embodiment of the invention can execute the method steps in the method embodiment, and the realization principle and the technical effect are similar, so that the details are not repeated.

It should be noted that the division of the modules of the above apparatus is only a logical division, and the actual implementation may be wholly or partially integrated into one physical entity, or may be physically separated. And these modules can all be implemented in the form of software invoked by a processing element; or may be implemented entirely in hardware; and part of the modules can be realized in the form of calling software by the processing element, and part of the modules can be realized in the form of hardware. For example, the obtaining module may be a processing element separately set up, or may be integrated into a chip of the apparatus, or may be stored in a memory of the apparatus in the form of program code, and a processing element of the apparatus calls and executes the function of the determining module. The other modules are implemented similarly. In addition, all or part of the modules can be integrated together or can be independently realized. The processing element described herein may be an integrated circuit having signal processing capabilities. In implementation, each step of the above method or each module above may be implemented by an integrated logic circuit of hardware in a processor element or an instruction in the form of software.

For example, the above modules may be one or more integrated circuits configured to implement the above methods, such as: one or more Application Specific Integrated Circuits (ASICs), or one or more Digital Signal Processors (DSPs), or one or more Field Programmable Gate Arrays (FPGAs), etc. For another example, when some of the above modules are implemented in the form of a Processing element scheduler code, the Processing element may be a general-purpose processor, such as a Central Processing Unit (CPU) or other processor that can invoke the program code. As another example, these modules may be integrated together and implemented in the form of a System-on-a-chip (SOC).

In the above embodiments, the implementation may be wholly or partially realized by software, hardware, firmware, or any combination thereof. When implemented in software, may be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. The procedures or functions described in accordance with the embodiments of the invention may be carried out in whole or in part when the computer program instructions are loaded and executed on a computer. The computer may be a general purpose computer, a special purpose computer, a network of computers, or other programmable device. The computer instructions may be stored on a computer readable storage medium or transmitted from one computer readable storage medium to another computer readable storage medium, for example, the computer instructions may be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wire (e.g., coaxial cable, fiber optic, Digital Subscriber Line (DSL)) or wireless (e.g., infrared, wireless, bluetooth, microwave, etc.). The computer-readable storage medium may be any available medium that can be accessed by a computer or a data storage device, such as a server, a data center, etc., that includes one or more available media. The usable medium may be a magnetic medium (e.g., a floppy disk, a hard disk, a magnetic tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a Solid State Disk (SSD)), etc.

Fig. 3 is a schematic structural diagram of an electronic device according to a third embodiment of the present invention. The electronic device may be the terminal device or the server, or may be a terminal device or a server connected to the terminal device or the server and implementing the method according to the embodiment of the present invention. As shown in fig. 3, the electronic device may include: a processor 31 (e.g., CPU), a memory 32, a transceiver 33; the transceiver 33 is coupled to the processor 31, and the processor 31 controls the transceiving operation of the transceiver 33. Various instructions may be stored in memory 32 for performing various processing functions and implementing the methods and processes provided in the above-described embodiments of the present invention. Preferably, the electronic device according to an embodiment of the present invention further includes: a power supply 34, a system bus 35, and a communication port 36. The system bus 35 is used to implement communication connections between the elements. The communication port 36 is used for connection communication between the electronic device and other peripherals.

The system bus mentioned in fig. 3 may be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus or the like. The system bus may be divided into an address bus, a data bus, a control bus, and the like. For ease of illustration, only one thick line is shown, but this does not mean that there is only one bus or one type of bus. The communication interface is used for realizing communication between the database access device and other equipment (such as a client, a read-write library and a read-only library). The Memory may include a Random Access Memory (RAM), and may further include a Non-Volatile Memory (Non-Volatile Memory), such as at least one disk Memory.

The Processor may be a general-purpose Processor, including a central processing unit CPU, a Network Processor (NP), and the like; but also a digital signal processor DSP, an application specific integrated circuit ASIC, a field programmable gate array FPGA or other programmable logic device, discrete gate or transistor logic, discrete hardware components.

It should be noted that the embodiment of the present invention also provides a computer-readable storage medium, which stores instructions that, when executed on a computer, cause the computer to execute the method and the processing procedure provided in the above-mentioned embodiment.

The embodiment of the invention also provides a chip for running the instructions, and the chip is used for executing the method and the processing process provided by the embodiment.

The embodiment of the invention provides a method, a device, electronic equipment and a computer readable storage medium for intelligently customizing a customer-obtaining publicity, which refer to an NLP technology to perform feature recognition on a publicity to be evaluated to obtain a corresponding text feature vector, perform feature recognition on commercial activity scene information corresponding to the publicity to obtain a corresponding scene feature vector, and then use a publicity effect recognition model to perform effect value prediction on the text feature vector and the scene feature vector, so that qualified or unqualified effect evaluation can be quickly given to each publicity to be selected on the premise of not depending on any artificial experience. By the method and the device, the degree of dependence on manual experience in the quality control process of the customized passenger publicity words is reduced, and the customization efficiency and quality of the passenger publicity words are improved.

The steps of a method or algorithm described in connection with the embodiments disclosed herein may be embodied in hardware, a software module executed by a processor, or a combination of the two. A software module may reside in Random Access Memory (RAM), memory, Read Only Memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art.

The above-mentioned embodiments, objects, technical solutions and advantages of the present invention are further described in detail, it should be understood that the above-mentioned embodiments are only examples of the present invention, and are not intended to limit the scope of the present invention, and any modifications, equivalent substitutions, improvements and the like made within the spirit and principle of the present invention should be included in the scope of the present invention.

Claims

1. A method for intelligently customizing a guest advertisement, the method comprising:

when the first passenger obtaining propaganda data is evaluated to be qualified propaganda data, outputting the first passenger obtaining propaganda data as the customized passenger obtaining propaganda data;

the text feature vector conversion processing is performed on the first passenger publicity language data to generate a corresponding first publicity text feature vector, and the method specifically includes:

performing word segmentation and part-of-speech tagging on each first sentence data based on a preset intelligent word segmentation model to generate a corresponding first word segmentation data set sequence, and forming a first word segmentation data set sequence set by all the obtained first word segmentation data set sequences; the first sequence of partial word data sets comprises a plurality of first partial word data sets; the first word segmentation data group comprises first word segmentation text data and first word segmentation data;

2. The method of claim 1, wherein the performing a textual syntax evaluation on the first sequence of component data sets to generate corresponding first syntax evaluation data comprises:

counting the number of the first branch word data group sequences in the first branch word data group sequence set to generate a first sentence total number; counting the number of the first sentence grammar evaluation data which are qualified to generate a first qualified sentence total number; generating a first qualified rate according to the ratio of the total number of the first qualified sentences to the total number of the first sentences;

3. The method according to claim 1, wherein the step of performing text style confirmation processing on the first activity scene data to generate corresponding first style data specifically comprises:

extracting a first product text style range field of a first product style data record of which a first product type field is matched with the first product type data of the first activity scene data from a preset first product style database to serve as a corresponding first text style range; the first product style database comprises a plurality of the first product style data records; the first product style data record comprises the first product type field and the first product text style range field; the first product text style range field comprises a plurality of text style types;

extracting a first activity text style range field of a first activity style data record of which the first activity type field is matched with the first activity type data of the first activity scene data from a preset first activity style database as a corresponding second text style range; the first activity style database comprises a plurality of the first activity style data records; the first activity style data record comprises the first activity type field and the first activity text style range field; the first active text style range field comprises a plurality of text style types;

4. The method according to claim 1, wherein the step of performing text style matching degree calculation processing on the first segmentation data group sequence set according to the first style data to generate corresponding first style matching degree data specifically comprises:

5. The method according to claim 1, wherein the performing scene keyword weight calculation processing on the first sequence of keyword sets according to the first activity scene data to generate corresponding first scene keyword weight data specifically comprises:

forming a third key word range by obtaining all the third key word data groups;

and performing sum calculation on all the obtained first keyword weight data to generate first scene keyword weight data.

6. The method of claim 1, wherein prior to using the trained promotional effect recognition model, the method further comprises training the promotional effect recognition model, specifically:

acquiring a plurality of groups of first historical data groups; the first historical data group comprises first historical propaganda language data, first historical activity scene data and first historical propaganda effect data; the first historical activity scenario data comprises second product type data and second activity type data; the first historical promotional effect data comprises a first campaign page click total amount, a first campaign participant total amount and a first campaign product sale total amount;

error calculation is carried out on the first training effect value data and the first historical effect value data of the corresponding first training data set based on a preset model error calculation function, and first effect value error data are generated;

when the first effect value error data is not in a preset reasonable effect value error range, modulating the neural network parameters and/or the neural network structure of the propaganda effect recognition model according to the first effect value error data, and continuing to train the propaganda effect recognition model by using the next group of first training data after modulation;

7. An apparatus for implementing the method of intelligently customizing a captured poster of any of claims 1-6, the apparatus comprising: the system comprises an acquisition module, a characteristic vector preparation module, a propaganda effect prediction module, a propaganda data qualification judgment module and a propaganda word customization output module;

and the propaganda language customization output module is used for outputting the first guest obtaining propaganda language data as the current customized guest obtaining propaganda language when the first guest obtaining propaganda language data are evaluated to be qualified propaganda data.

8. An electronic device, comprising: a memory, a processor, and a transceiver;

the processor is used for being coupled with the memory, reading and executing the instructions in the memory to realize the method steps of any one of claims 1-6;

9. A computer-readable storage medium having computer instructions stored thereon which, when executed by a computer, cause the computer to perform the method of any of claims 1-6.