CN116228280A - User demand prediction method based on big data - Google Patents
User demand prediction method based on big data Download PDFInfo
- Publication number
- CN116228280A CN116228280A CN202310302424.2A CN202310302424A CN116228280A CN 116228280 A CN116228280 A CN 116228280A CN 202310302424 A CN202310302424 A CN 202310302424A CN 116228280 A CN116228280 A CN 116228280A
- Authority
- CN
- China
- Prior art keywords
- marketing
- data
- input
- time sequence
- convolution
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q30/00—Commerce
- G06Q30/02—Marketing; Price estimation or determination; Fundraising
- G06Q30/0201—Market modelling; Market analysis; Collecting market data
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q30/00—Commerce
- G06Q30/02—Marketing; Price estimation or determination; Fundraising
- G06Q30/0201—Market modelling; Market analysis; Collecting market data
- G06Q30/0202—Market predictions or forecasting for commercial activities
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q30/00—Commerce
- G06Q30/02—Marketing; Price estimation or determination; Fundraising
- G06Q30/0241—Advertisements
- G06Q30/0242—Determining effectiveness of advertisements
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q30/00—Commerce
- G06Q30/02—Marketing; Price estimation or determination; Fundraising
- G06Q30/0241—Advertisements
- G06Q30/0277—Online advertisement
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management
Landscapes
- Business, Economics & Management (AREA)
- Engineering & Computer Science (AREA)
- Finance (AREA)
- Accounting & Taxation (AREA)
- Development Economics (AREA)
- Strategic Management (AREA)
- Theoretical Computer Science (AREA)
- Entrepreneurship & Innovation (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Game Theory and Decision Science (AREA)
- Economics (AREA)
- Marketing (AREA)
- General Business, Economics & Management (AREA)
- Data Mining & Analysis (AREA)
- Artificial Intelligence (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The user demand prediction predicts the future demands of users by analyzing factors such as user behaviors, market trends, competitors and the like, is beneficial to enterprises to better formulate market strategies in aspects of product research and development, marketing, pricing and the like, and better knows market trends and user demands, so that decision risks are reduced. The arrival of big data age and the increasingly personalized diversification of user demands promote the prediction of big data user demands to become an important tool for enterprise decision and marketing. Therefore, the invention provides the user demand prediction method based on big data, which comprehensively considers sales promotion marketing activities, historical demands and other factors to realize accurate demand prediction and helps enterprises to make more intelligent decisions in production, purchasing and selling aspects.
Description
Technical Field
The invention belongs to the field of user demand prediction, and particularly relates to a user demand prediction method based on big data.
Background
The user demand prediction predicts the future demands of users by analyzing factors such as user behaviors, market trends, competitors and the like, is beneficial to enterprises to better formulate market strategies in aspects of product research and development, marketing, pricing and the like, and better knows market trends and user demands, so that decision risks are reduced. The existing user demand prediction method mainly comprises the following four types: 1) And (5) market research. Acquiring data by means of questionnaires, focus groups, user interviews and the like, so as to predict future user demands; 2) Competition analysis. Through analyzing the products and market strategies of competitors, the trend of the market and the user demand can be known, so that the future user demand can be predicted; 3) Social media analysis. By analyzing the discussion and feedback of the users on the social media, the future user demands can be predicted; 4) Expert opinion. Expert opinion can provide valuable market information and predictions to analyze market trends and user needs from an industry perspective. With the development of technologies such as the internet, the mobile internet and the internet of things, the scale and the variety of data are explosively increased. However, the advent of the big data age and the increasing personalized diversification of user needs have prompted big data user needs prediction to become an important tool for enterprise decision making and marketing. On the other hand. The development and technical support of the prediction model enable the prediction of the demand of big data users to be possible, and can help enterprises to better grasp market opportunities and optimize business processes, and promote competitiveness and benefit.
Therefore, the invention provides the user demand prediction method based on big data, which comprehensively considers sales promotion marketing activities, historical demands and other factors to realize accurate demand prediction and helps enterprises to make more intelligent decisions in production, purchasing and selling aspects.
Disclosure of Invention
The invention aims to overcome the defects in the prior art and provides a user demand prediction method based on big data, which adopts the following technical scheme:
s1, for a product to be predicted, acquiring marketing input time sequence data of the product in 5 channels of television advertisements, short messages, network push messages, advertisement mails and network news, acquiring sales volume and sales profits, and performing data complement and data conversion on the acquired 7 types of time sequence data;
s2, calculating marketing effects of the advertising media by using a marketing combination model, calculating accumulated influence effects of different marketing activities and extracting advertising marketing features;
s202, calculating contribution of advertising marketing activities according to the marketing input time sequence data acquired in the step S1
S204, extracting advertisement marketing characteristics according to the contribution of the promotion marketing activities in S202
S3, analyzing a time sequence variation trend of the user demand, performing causal convolution on the sales time sequence data acquired in the step S1, the consumer confidence index, the producer price index and the consumer price index by using a time sequence convolution network, and extracting time sequence characteristics of the user demand;
s4, merging the user demand time sequence features extracted in the step S3 with the advertisement marketing features extracted in the step S2, and predicting the user demands;
s402, calculating weights of different features by using an attention mechanism, and applying the attention weights to the time sequence features and the advertisement marketing features required by the user to obtain final feature representation
S404, carrying out regression prediction of the end user demands by utilizing a two-layer fully connected network
Preferably, the data acquisition, completion and conversion in step S1 are specifically:
time series data required for user demand prediction is collected from the internet, and the data includes marketing investment data and time series data of sales. The marketing data comprises sales volume and sales profits, and the marketing investment data comprises investment of 5 marketing channels such as television advertisements, short messages, network push messages, advertisement mails, network news and the like.
For missing values in the time series data, the following processing strategies are adopted: 1) Filling missing data by using the mean value and the median as estimated values; 2) When data is available only when a transaction or promotion occurs during the day, the lost data may simply be replaced with zero to indicate that there is no transaction or promotion for the day; 3) And if the missing values are more, deleting the missing data.
In order to fit the reality situation, consider the attenuation effects of television advertisements, short messages, network push messages, advertisement mails and network news, namely the delay effect of advertisements:
A t ＝A t +r*A t-1
wherein A is t Indicating the impact of advertising marketing at time t, A t-1 The impact of advertising marketing at time t-1 is represented, and r is the delay factor of advertising impact, which takes a value between 0 and 1.
In order to fit the actual diminishing returns effect, the marketing investment time sequence data is subjected to data conversion, and the data conversion is calculated as follows:
where Y represents data after conversion, x represents data before conversion, and a, b, c, and d are 4 parameters whose settings are set according to a specific marketing type.
Preferably, the contribution of the advertising marketing campaign in step S202 is specifically:
in order to measure the current influence contribution of different marketing modes to consumers, the current influence of a persistent marketing campaign needs to be calculated, namely the weight factor of the marketing input history time sequence data at the current moment, and the calculation mode is as follows:
wherein t represents the current time, t i For a historical time, d represents the decay factor of the advertising effect,
indicating time t i Residual marketing effectiveness weight of marketing campaign to time t.
The cumulative marketing effect at the current time t is calculated as:
wherein the method comprises the steps of
Representing the contribution of the marketing campaign at the current moment t +.>
Indicating time t i Contribution of marketing campaign->
Indicating time t i The attenuated current weight of the contribution of the marketing campaign.
Preferably, the advertisement cumulative effect evaluation model in step S204 specifically includes:
designing an advertisement cumulative effect evaluation model based on a fuzzy neural network, wherein the first layer of the model is an input layer, and the input data is the cumulative marketing effect of the current moment calculated in S202
The fuzzy function of the second layer is Gaussian activation function, and the third layer is output layer.
In the fuzzy neural network, the input characteristics are converted into fuzzy variables through membership functions, then the fuzzy variables are input into neurons, the output of the neurons is subjected to inverse membership functions to obtain fuzzy sets, and the final output characteristics are obtained through synthesis or fuzzy reasoning of the fuzzy sets.
The blur function of the second layer uses a gaussian activation function, which is calculated as follows:
where x represents the network input, i.e. the marketing campaign contribution in the previous step, a is the mean value, σ is the standard deviation of the fuzzy function, exp (·) represents the exponential function.
The output formula of the output layer, for the jth output dimension, the output can be expressed as follows:
wherein w is ij Is the weight, mu, connecting the hidden layer i-th neuron and the output layer j-th neuron i Is the fuzzy membership corresponding to the input characteristic and is calculated by the fuzzy function.
Calculating the fuzzy neural network: 1) Generating weights w for input and output hidden layers, where each w ji Are fuzzy numbers; 2) Assigning values for alpha and eta for training fuzzy back propagation; 3) Acquiring a next mode set, and performing hidden calculation on the manual input and output neurons; 4) Calculating weight changes of the input hidden layer and the output hidden layer; 5) Updating the weight of the input hidden layer; 6) Updating the weight of the hidden output layer; 7) The outputs w' and w "of the final fuzzy membership weight set are calculated.
Preferably, the analysis of the time sequence change of the user demand in the step S3 specifically includes:
the time sequence convolution network adopts a one-dimensional full convolution structure, and the input and the output of each layer of convolution are equal in length. The time sequence convolution is a causal convolution, namely, the output of the current time point t only depends on the data between the time point t and the time point t, so that the leakage of future data is avoided, and the time sequence convolution is suitable for a scene of time sequence prediction.
The input of the time sequence convolution network comprises 5 characteristics of sales volume and sales profits obtained in the step S1, and the acquired consumer confidence index, the producer price index and the consumer price index form an input sequence of 5*n, wherein n represents the length of the time sequence and is multichannel input.
The process of the time sequence convolution is as follows:
where seq represents an input time sequence, i.e. filter is a convolution kernel, k represents the kth channel of the output, seq (c, w) represents the w-th element of the input c-th channel, and filter (k, c, s) represents the s-th element of the convolution kernel corresponding to the c-th input channel, the kth output channel.
In order to further expand the convolution field of view without increasing the computational complexity, a hole convolution parameter d is introduced, i.e., the output layer data point t is obtained by convolution of the input layer points t, t-d, t-2d, the formalized formula is:
preferably, the feature fusion in step S402 is specifically:
for each feature x i Its attention weight a can be calculated i The weight represents the importance of the feature to the final prediction result. The specific calculation mode can be realized by using a feedforward neural network, wherein the input of the network is the original characteristic, and the output is the attention weight. Wherein the weights may be normalized using a softmax function to ensure that their sum is 1, i.e.:
wherein z is i The output of the ith feature is represented, n represents the number of features, exp (·) represents the exponential function.
For each feature x i It can be multiplied by the corresponding attention weight a i The products of all features are then added to obtain the final feature representation z, namely:
where n represents the number of features where n=2, i.e. includes both historical user demand timing features and advertising marketing features.
Preferably, step S404 specifically includes:
for the feature representation z, its feature dimension n, then the regression layer is made up of a two-layer network, which can be represented as:
y＝σ(W 2 σ(W 1 z+b 1 )+b2)
wherein W is 1 ∈R h×n And W is 2 ∈R 1×h Is two weight matrices, b 1 And b2 are two bias vectors, h is the hidden layer dimension in the network, σ is the ReLU activation function to ensure that the end user's demand prediction is non-negative.
And obtaining a predicted value of the end user demand through a regression layer.
The invention has the beneficial effects that: the present invention takes into account the advertising effectiveness and contribution of the promotional marketing campaign, and for the prediction manager, the present invention can be used to optimize revenue, process large data sets, improve prediction accuracy, calculate the impact of advertising on sales, link demand and offer data, and minimize marketing investment based on the promotional campaign and other marketing factors.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below. The drawings in the following description are only examples of embodiments of the present invention and other drawings may be made from these drawings by those of ordinary skill in the art without undue burden.
Wherein:
FIG. 1 is a drawing of the abstract of the specification of the present invention;
FIG. 2 is a flow chart of advertisement marketing feature extraction in an embodiment of the present invention;
FIG. 3 is a flow chart of feature fusion and user demand prediction in an embodiment of the present invention;
Detailed Description
The following description of the embodiments of the present invention will be made clearly and perfectly with reference to the accompanying drawings of the embodiments of the present invention, and it is apparent that the described embodiments are only some embodiments of the present invention, not all embodiments. All other embodiments, which are obtained without inventive effort by a person skilled in the art based on the embodiments of the present invention, are within the scope of the present invention.
Example 1
The user demand prediction method provided by one embodiment of the invention comprises the following steps:
s1, for a product to be predicted, acquiring marketing input time sequence data of the product in 5 channels of television advertisements, short messages, network push messages, advertisement mails and network news, acquiring sales volume and sales profits, and performing data complement and data conversion on the acquired 7 types of time sequence data;
s2, calculating marketing effects of the advertising media by using a marketing combination model, calculating accumulated influence effects of different marketing activities and extracting advertising marketing features;
s202, calculating contribution of advertising marketing activities according to the marketing input time sequence data acquired in the step S1
S204, extracting advertisement marketing characteristics according to the contribution of the promotion marketing activities in S202
S3, analyzing a time sequence variation trend of the user demand, performing causal convolution on the sales time sequence data acquired in the step S1, the consumer confidence index, the producer price index and the consumer price index by using a time sequence convolution network, and extracting time sequence characteristics of the user demand;
s4, merging the user demand time sequence features extracted in the step S3 with the advertisement marketing features extracted in the step S2, and predicting the user demands;
s402, calculating weights of different features by using an attention mechanism, and applying the attention weights to the time sequence features and the advertisement marketing features required by the user to obtain final feature representation
S404, carrying out regression prediction of the end user demands by utilizing a two-layer fully connected network
Example two
On the basis of the first embodiment, the specific steps of data acquisition, complementation and conversion are as follows: time series data required for user demand prediction is collected from the internet, and the data includes marketing investment data and time series data of sales. The marketing data comprises sales volume and sales profits, and the marketing investment data comprises investment of 5 marketing channels such as television advertisements, short messages, network push messages, advertisement mails, network news and the like.
For missing values in the time series data, the following processing strategies are adopted: 1) Filling missing data by using the mean value and the median as estimated values; 2) When data is available only when a transaction or promotion occurs during the day, the lost data may simply be replaced with zero to indicate that there is no transaction or promotion for the day; 3) And if the missing values are more, deleting the missing data.
In order to fit the reality situation, consider the attenuation effects of television advertisements, short messages, network push messages, advertisement mails and network news, namely the delay effect of advertisements:
A t ＝A t +r*A t-1
wherein A is t Indicating the impact of advertising marketing at time t, A t-1 The impact of advertising marketing at time t-1 is represented, and r is the delay factor of advertising impact, which takes a value between 0 and 1.
In order to fit the actual diminishing returns effect, the marketing investment time sequence data is subjected to data conversion, and the data conversion is calculated as follows:
where Y represents data after conversion, x represents data before conversion, and a, b, c, and d are 4 parameters whose settings are set according to a specific marketing type.
Example III
On the basis of the first and second embodiments, as shown in fig. 2, the contribution of the advertising marketing campaign in the step S202 of the method of the present invention is specifically:
in order to measure the current influence contribution of different marketing modes to consumers, the current influence of a persistent marketing campaign needs to be calculated, namely the weight factor of the marketing input history time sequence data at the current moment, and the calculation mode is as follows:
wherein t represents the current time, t i For a historical time, d represents the decay factor of the advertising effect,
indicating time t i Residual marketing effectiveness weight of marketing campaign to time t.
The cumulative marketing effect at the current time t is calculated as:
wherein the method comprises the steps of
Representing the contribution of the marketing campaign at the current moment t +.>
Indicating time t i Contribution of marketing campaign->
Indicating time t i The attenuated current weight of the contribution of the marketing campaign.
The advertisement cumulative effect evaluation model in step S204 of the method specifically includes:
designing an advertisement cumulative effect evaluation model based on a fuzzy neural network, wherein the first layer of the model is an input layer, and the input data is the cumulative marketing effect of the current moment calculated in S202
The fuzzy function of the second layer is Gaussian activation function, and the third layer is output layer.
In the fuzzy neural network, the input characteristics are converted into fuzzy variables through membership functions, then the fuzzy variables are input into neurons, the output of the neurons is subjected to inverse membership functions to obtain fuzzy sets, and the final output characteristics are obtained through synthesis or fuzzy reasoning of the fuzzy sets.
The blur function of the second layer uses a gaussian activation function, which is calculated as follows:
where x represents the network input, i.e. the marketing campaign contribution in the previous step, a is the mean value, σ is the standard deviation of the fuzzy function, exp (·) represents the exponential function.
The output formula of the output layer, for the jth output dimension, the output can be expressed as follows:
wherein w is ij Is the weight, mu, connecting the hidden layer i-th neuron and the output layer j-th neuron i Is the fuzzy membership corresponding to the input characteristic and is calculated by the fuzzy function.
Calculating the fuzzy neural network: 1) Generating weights w for input and output hidden layers, where each w ji Are fuzzy numbers; 2) Assigning values for alpha and eta for training fuzzy back propagation; 3) Acquiring a next mode set, and performing hidden calculation on the manual input and output neurons; 4) Calculating weight changes of the input hidden layer and the output hidden layer; 5) Updating the weight of the input hidden layer; 6) Updating the weight of the hidden output layer; 7) The outputs w' and w "of the final fuzzy membership weight set are calculated.
Example IV
The time sequence change analysis of the user demand in the step S3 of the method specifically comprises the following steps:
the time sequence convolution network adopts a one-dimensional full convolution structure, and the input and the output of each layer of convolution are equal in length. The time sequence convolution is a causal convolution, namely, the output of the current time point t only depends on the data between the time point t and the time point t, so that the leakage of future data is avoided, and the time sequence convolution is suitable for a scene of time sequence prediction.
The input of the time sequence convolution network comprises 5 characteristics of sales volume and sales profits obtained in the step S1, and the acquired consumer confidence index, the producer price index and the consumer price index form an input sequence of 5*n, wherein n represents the length of the time sequence and is multichannel input.
The process of the time sequence convolution is as follows:
where seq represents an input time sequence, i.e. filter is a convolution kernel, k represents the kth channel of the output, seq (c, w) represents the w-th element of the input c-th channel, and filter (k, c, s) represents the s-th element of the convolution kernel corresponding to the c-th input channel, the kth output channel.
In order to further expand the convolution field of view without increasing the computational complexity, a hole convolution parameter d is introduced, i.e., the output layer data point t is obtained by convolution of the input layer points t, t-d, t-2d, the formalized formula is:
example five
As shown in fig. 3, in the embodiment of the present invention, the specific content of fusing features and predicting user demands includes:
for each feature x i Its attention weight a can be calculated i The weight represents the importance of the feature to the final prediction result. The specific calculation mode can be realized by using a feedforward neural network, wherein the input of the network is the original characteristic, and the output is the attention weight. Wherein the weights may be normalized using a softmax function to ensure that their sum is 1, i.e.:
wherein z is i The output of the ith feature is represented, n represents the number of features, exp (·) represents the exponential function.
For each feature x i It can be multiplied by the corresponding attention weight a i The products of all features are then added to obtain the final feature representation z, namely:
where n represents the number of features where n=2, i.e. includes both historical user demand timing features and advertising marketing features. For the feature representation z, its feature dimension n, then the regression layer is made up of a two-layer network, which can be represented as:
y＝σ(W 2 σ(W 1 z+b 1 )+b2)
wherein W is 1 ∈R h×n And W is 2 ∈R 1×h Is two weight matrices, b 1 And b2 are two bias vectors, h is the hidden layer dimension in the network, σ is the ReLU activation function to ensure that the end user's demand prediction is non-negative.
And obtaining a predicted value of the end user demand through a regression layer.
In summary, the embodiment of the invention provides a new market demand prediction to comprehensively consider the historical demand, improve the prediction accuracy, and consider the marketing campaign and the decline of the marketing effect thereof along with time so as to consider the reality factors as much as possible and fit the actual application scene. According to the method, feature extraction is carried out on time-sequence market demands and marketing activities based on time attenuation, a time sequence convolution network and a fuzzy neural network, feature fusion is carried out through an attention mechanism, the predicted market demands are finally obtained, and the accuracy of market demand prediction is greatly improved.
Experiment verification
In order to verify the effectiveness of the user demand prediction method based on big data, the invention performs performance test under the following 5 scenes based on sales information of a real product:
scene 1: adjustments to advertising expenditures and other activities are not considered and are assumed to be the same as in the last 2015.
Scene 2: considering the 5% rise in unit price, the advertising expenditure for different marketing channels rises by 10%.
Scene 3: television advertising expenditure increased dramatically by 10%, all other marketing channels decreased by 5% and unit price did not rise.
Scene 4: considering the unit price and television advertising spending increased by 10%, all other marketing channel spending decreased by 10%.
Scene 5: considering the increase in unit price by 10%, advertising costs for television and flat media are both increased, and costs for all other marketing channels are reduced by 10%.
TABLE 1 user demand prediction error
| Scene 1 | Scene 2 | Scene 3 | Scene 4 | Scene 5 | |
| Predicting demand values | 112465 | 156569 | 756871 | 1567997 | 896691 |
| True demand value | 103275 | 167854 | 712256 | 1489755 | 967864 |
| Prediction error | 8.18％ | 7.21％ | 5.88％ | 4.98％ | 7.93％ |
The prediction results of the invention are shown in table 1, and the prediction accuracy of the invention is higher than 91% in different scenes, so that a good effect is obtained.
The above examples only represent a few embodiments of the present application, which are described in more detail and are not to be construed as limiting the scope of the present application. It should be noted that it would be apparent to those skilled in the art that various modifications and improvements could be made without departing from the spirit of the present application, which would be within the scope of the present application. Accordingly, the scope of protection of the present application is to be determined by the claims appended hereto.
Claims (6)
1. The user demand prediction method based on big data is characterized by comprising the following steps:
s1, for a product to be predicted, acquiring marketing input time sequence data of the product in 5 channels of television advertisements, short messages, network push messages, advertisement mails and network news, acquiring sales volume and sales profits, and performing data complement and data conversion on the acquired 7 types of time sequence data;
s2, calculating marketing effects of the advertising media by using a marketing combination model, calculating accumulated influence effects of different marketing activities and extracting advertising marketing features;
s202, calculating contribution of advertising marketing activities according to the marketing input time sequence data acquired in the step S1
S204, extracting advertisement marketing characteristics according to the contribution of the promotion marketing activities in S202
S3, analyzing a time sequence variation trend of the user demand, performing causal convolution on the sales time sequence data acquired in the step S1, the consumer confidence index, the producer price index and the consumer price index by using a time sequence convolution network, and extracting time sequence characteristics of the user demand;
s4, merging the user demand time sequence features extracted in the step S3 with the advertisement marketing features extracted in the step S2, and predicting the user demands;
s402, calculating weights of different features by using an attention mechanism, and applying the attention weights to the time sequence features and the advertisement marketing features required by the user to obtain final feature representation
S404, carrying out regression prediction of the end user demands by utilizing a two-layer fully connected network.
2. The big data based user demand prediction method as claimed in claim 1, wherein the slave data collection, completion and conversion of S102 comprises:
time series data required for user demand prediction is collected from the internet, and the data includes marketing investment data and time series data of sales. The marketing data comprises sales volume and sales profits, and the marketing investment data comprises investment of 5 marketing channels such as television advertisements, short messages, network push messages, advertisement mails, network news and the like.
For missing values in the time series data, the following processing strategies are adopted: 1) Filling missing data by using the mean value and the median as estimated values; 2) When data is available only when a transaction or promotion occurs during the day, the lost data may simply be replaced with zero to indicate that there is no transaction or promotion for the day; 3) And if the missing values are more, deleting the missing data.
In order to fit the reality situation, consider the attenuation effects of television advertisements, short messages, network push messages, advertisement mails and network news, namely the delay effect of advertisements:
A t ＝A t +r*A t-1
wherein A is t Indicating the impact of advertising marketing at time t, A t-1 The impact of advertising marketing at time t-1 is represented, and r is the delay factor of advertising impact, which takes a value between 0 and 1.
In order to fit the actual diminishing returns effect, the marketing investment time sequence data is subjected to data conversion, and the data conversion is calculated as follows:
where Y represents data after conversion, x represents data before conversion, and a, b, c, and d are 4 parameters whose settings are set according to a specific marketing type.
3. The method for predicting user demand based on big data as claimed in claim 1, wherein the data conversion and sequence decomposition in S202 comprises:
in order to measure the current influence contribution of different marketing modes to consumers, the current influence of a persistent marketing campaign needs to be calculated, namely the weight factor of the marketing input history time sequence data at the current moment, and the calculation mode is as follows:
wherein t represents the current time, t i For a historical time, d represents the decay factor of the advertising effect,
indicating time t i Residual marketing effectiveness weight of marketing campaign to time t.
The cumulative marketing effect at the current time t is calculated as:
wherein the method comprises the steps of
Representing the contribution of the marketing campaign at the current moment t +.>
Indicating time t i Contribution of marketing campaign->
Indicating time t i The attenuated current weight of the contribution of the marketing campaign.
4. The method for predicting user demand based on big data as claimed in claim 1, wherein the calculating of the advertisement accumulation effect in S204 comprises:
designing an advertisement cumulative effect evaluation model based on a fuzzy neural network, wherein the first layer of the model is an input layer, and the input data is the cumulative marketing effect of the current moment calculated in S202
The fuzzy function of the second layer is Gaussian activation function, and the third layer is output layer.
In the fuzzy neural network, the input characteristics are converted into fuzzy variables through membership functions, then the fuzzy variables are input into neurons, the output of the neurons is subjected to inverse membership functions to obtain fuzzy sets, and the final output characteristics are obtained through synthesis or fuzzy reasoning of the fuzzy sets.
The blur function of the second layer uses a gaussian activation function, which is calculated as follows:
where x represents the network input, i.e. the marketing campaign contribution in the previous step, a is the mean value, σ is the standard deviation of the fuzzy function, exp (·) represents the exponential function.
The output formula of the output layer, for the jth output dimension, the output can be expressed as follows:
wherein w is ij Is the weight, mu, connecting the hidden layer i-th neuron and the output layer j-th neuron i Is the fuzzy membership corresponding to the input characteristic and is calculated by the fuzzy function.
5. The method for predicting user demand based on big data as claimed in claim 1, wherein the extracting of the time sequence features of the user demand in S3 comprises:
the time sequence convolution network adopts a one-dimensional full convolution structure, and the input and the output of each layer of convolution are equal in length. The time sequence convolution is a causal convolution, namely, the output of the current time point t only depends on the data between the time point t and the time point t, so that the leakage of future data is avoided, and the time sequence convolution is suitable for a scene of time sequence prediction.
The input of the time sequence convolution network comprises 5 characteristics of sales volume and sales profits obtained in the step S1, and the acquired consumer confidence index, the producer price index and the consumer price index form an input sequence of 5*n, wherein n represents the length of the time sequence and is multichannel input.
The process of the time sequence convolution is as follows:
where seq represents an input time sequence, i.e. filter is a convolution kernel, k represents the kth channel of the output, seq (c, w) represents the w-th element of the input c-th channel, and filter (k, c, s) represents the s-th element of the convolution kernel corresponding to the c-th input channel, the kth output channel.
In order to further expand the convolution field of view without increasing the computational complexity, a hole convolution parameter d is introduced, i.e., the output layer data point t is obtained by convolution of the input layer points t, t-d, t-2d, the formalized formula is:
6. the method for predicting user demand based on big data as recited in claim 1, wherein the feature fusion based on the attention mechanism in S402 comprises:
for each feature x i Its attention weight a can be calculated i The weight represents the importance of the feature to the final prediction result. The specific calculation mode can be realized by using a feedforward neural network, wherein the input of the network is the original characteristic, and the output is the attention weight. Wherein the weights may be normalized using a softmax function to ensure that their sum is 1, i.e.:
wherein z is i The output of the ith feature is represented, n represents the number of features, exp (·) represents the exponential function.
For each feature x i It can be multiplied by the corresponding attention weight a i The products of all features are then added to obtain the final feature representation z, namely:
where n represents the number of features where n=2, i.e. includes both historical user demand timing features and advertising marketing features.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310302424.2A CN116228280A (en) | 2023-03-27 | 2023-03-27 | User demand prediction method based on big data |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310302424.2A CN116228280A (en) | 2023-03-27 | 2023-03-27 | User demand prediction method based on big data |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116228280A true CN116228280A (en) | 2023-06-06 |
Family
ID=86585672
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310302424.2A Pending CN116228280A (en) | 2023-03-27 | 2023-03-27 | User demand prediction method based on big data |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116228280A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117670413A (en) * | 2023-12-13 | 2024-03-08 | 中教畅享科技股份有限公司 | Market crowd behavior-based market prediction method |
-
2023
- 2023-03-27 CN CN202310302424.2A patent/CN116228280A/en active Pending
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117670413A (en) * | 2023-12-13 | 2024-03-08 | 中教畅享科技股份有限公司 | Market crowd behavior-based market prediction method |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108648049B (en) | Sequence recommendation method based on user behavior difference modeling | |
| CN109509033B (en) | Big data prediction method for user purchasing behavior in consumption financial scene | |
| CN109284866B (en) | Commodity order prediction method and device, storage medium and terminal | |
| CN111008858A (en) | Commodity sales prediction method and system | |
| CN110796313B (en) | Session recommendation method based on weighted graph volume and item attraction model | |
| CN111160968A (en) | SKU-level commodity sales prediction method and device | |
| CN112365283B (en) | Coupon issuing method and device, terminal equipment and storage medium | |
| CN112163963A (en) | Service recommendation method and device, computer equipment and storage medium | |
| US11861654B2 (en) | System and method for predicting customer behavior | |
| CN112149352B (en) | Prediction method for marketing activity clicking by combining GBDT automatic characteristic engineering | |
| CN111652735A (en) | Insurance product recommendation method based on user behavior label characteristics and commodity characteristics | |
| CN116228280A (en) | User demand prediction method based on big data | |
| CN112650933A (en) | High-order aggregation-based graph convolution and multi-head attention mechanism conversation recommendation method | |
| CN115545886A (en) | Overdue risk identification method, overdue risk identification device, overdue risk identification equipment and storage medium | |
| Singha et al. | Application of different machine learning models for supply chain demand forecasting: comparative analysis | |
| CN115080868A (en) | Product pushing method, product pushing device, computer equipment, storage medium and program product | |
| CN111177581A (en) | Multi-platform-based social e-commerce website commodity recommendation method and device | |
| CN113850654A (en) | Training method of item recommendation model, item screening method, device and equipment | |
| CN113763095B (en) | Information recommendation method and device and model training method and device | |
| CN113610610A (en) | Session recommendation method and system based on graph neural network and comment similarity | |
| Kabir et al. | Integrating fuzzy Delphi method with artificial neural network for demand forecasting of power engineering company | |
| CN112232388A (en) | ELM-RFE-based shopping intention key factor identification method | |
| CN117131261A (en) | E-commerce content recommendation system based on AI and big data analysis | |
| CN116797235A (en) | Method and device for processing consumption information, storage medium and computer equipment | |
| CN111008861A (en) | Intelligent prediction-based commodity online transaction system and method |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |