CN112927037A - Supplier recommendation method and system - Google Patents

Supplier recommendation method and system Download PDF

Info

Publication number
CN112927037A
CN112927037A CN202110156145.0A CN202110156145A CN112927037A CN 112927037 A CN112927037 A CN 112927037A CN 202110156145 A CN202110156145 A CN 202110156145A CN 112927037 A CN112927037 A CN 112927037A
Authority
CN
China
Prior art keywords
supplier
recommendation
user
function
nonlinear
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.)
Granted
Application number
CN202110156145.0A
Other languages
Chinese (zh)
Other versions
CN112927037B (en
Inventor
李勇刚
李育东
阳春华
黄科科
朱红求
陈宇
刘卫平
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Central South University
Original Assignee
Central South University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Central South University filed Critical Central South University
Priority to CN202110156145.0A priority Critical patent/CN112927037B/en
Publication of CN112927037A publication Critical patent/CN112927037A/en
Application granted granted Critical
Publication of CN112927037B publication Critical patent/CN112927037B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

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
    • G06Q30/06—Buying, selling or leasing transactions
    • G06Q30/0601—Electronic shopping [e-shopping]
    • G06Q30/0605—Supply or demand aggregation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00—Machine learning
    • 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
    • Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
    • Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
    • Y02P90/30—Computing systems specially adapted for manufacturing

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Business, Economics & Management (AREA)
  • General Physics & Mathematics (AREA)
  • Accounting & Taxation (AREA)
  • Finance (AREA)
  • Software Systems (AREA)
  • Physics & Mathematics (AREA)
  • Marketing (AREA)
  • Evolutionary Computation (AREA)
  • Strategic Management (AREA)
  • Economics (AREA)
  • Artificial Intelligence (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Data Mining & Analysis (AREA)
  • General Business, Economics & Management (AREA)
  • Medical Informatics (AREA)
  • Development Economics (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Mathematical Physics (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention relates to the technical field of computers, and discloses a supplier recommendation method and a supplier recommendation system, which are used for processing a nonlinear recommendation problem along with the interest and feature change of a user and ensuring a recommendation effect. The method comprises the following steps: receiving at least one supplier sample; predicting the recommended operation corresponding to each supplier sample based on an adaptive multi-layer perception online transfer learning model; tracking and acquiring the actual adoption effect of the supplier user on each supplier sample; according to the comparison between the recommended operation and the actual adoption effect, the weight distribution and the nonlinear parameter vector between the user characteristics of the previous batch of the purchasing users served by the self-adaptive multi-layer perception online migration learning model and the user characteristics of the current batch are adjusted; and re-acquiring a new supplier sample, predicting corresponding recommended operation by using the adjusted online transfer learning model, and repeating the steps until all the supplier samples meeting the supplier users are screened out.

Description

Supplier recommendation method and system
Technical Field
The invention relates to the technical field of computers, in particular to a supplier recommendation method and a supplier recommendation system.
Background
The nonferrous metallurgy supplier selection problem is a multi-attribute decision-making problem, and a purchasing department of an enterprise selects a proper supplier to purchase according to the requirements of the enterprise, so that the benefit and the efficiency of the enterprise are maximized. As the enterprise evolves, the number of alternative suppliers for each procurement campaign increases, and new suppliers may be added. The characteristics of nonferrous metallurgy suppliers become complex with the improvement of production requirements, and the decision efficiency of purchasing departments is often reduced in the face of a large number of suppliers with complex information. Such problems can be attributed to vendor information overload problems. On the other hand, the demands of the purchasing department on the suppliers can change in the operation process of the nonferrous metallurgy enterprises. For non-ferrous metallurgy type enterprises which need to maintain continuous production, the characteristics of suppliers determine the production emphasis of enterprises in a period of time. In this period, the enterprise is influenced by factors such as working conditions and environment, the enterprise has different demands on suppliers at different times, and the demand changes rapidly and inevitably. For example, nonferrous metallurgy companies have many requirements for raw material suppliers, and in the case of stable production and inventory, the purchasing department will prefer the price factor. When the production conditions are unstable, the purchasing department will prefer to supply the suppliers with stable quality. In this case, the information overload of the supplier will affect the decision efficiency of the purchasing department, and the demand change of the purchasing department will make the recommendation of the supplier more complicated. In the face of such lack of awareness of resources by the purchasing department, the recommendation system can learn the needs of the purchasing department and generate appropriate supplier recommendations. Therefore, the establishment of the recommendation system which can adapt to the demand change of the purchasing department and can process the information overload of the suppliers has very important practical significance for the purchasing decision of the nonferrous metallurgy enterprises.
In the face of a large amount of project information,users need recommendation systems to improve the efficiency of decisions. FIG. 1 illustrates a training process for a recommendation system for changes in user demand. S represents the requirement characteristics of the user, and x and y represent the item recommended in a single batch and the user operation result (label), respectively. The recommendation system trains the user demand characteristics according to the operation results of each batch of purchasing departments and gradually obtains stable demand characteristics. The recommendation system then predicts the new item based on the trained demand characteristics to complete the recommendation. However, during use, the requirements of the user may change, and the recommendation system needs to be able to follow the requirements of the user to maintain the accuracy of the recommendation, which can be attributed to a concept drift problem. First, the user' S needs remain stable for an initial period of time, which means that the trained need characteristics S1＝S2＝…＝St+1S. For any item x at this timetRandom StPredicted result y of (2)tAre all substantially identical. Next, at some point j, the user' S actions change such that the trained user requirements S change, Si≠Sj。
However, the prior art has the following drawbacks:
1. the nonferrous metallurgy supplier recommendations are a non-linear classification problem, which means that it must be solved in a non-linear way.
2. In the operation process, the requirements of the purchasing department can be influenced by the outside world to change. The solution is to follow the change of the user characteristics by online learning, because the online learning only adjusts the user requirement characteristics according to the current sample, the instantaneity can ignore the sample conflict problem brought by the old data set. However, the online learned user characteristics fluctuate due to erroneous samples generated by possible user misoperations, which in turn leads to fluctuations in recommendation accuracy.
Disclosure of Invention
The invention aims to disclose a supplier recommendation method and a supplier recommendation system, which are used for processing a nonlinear recommendation problem along with the interest and feature change of a user and ensuring a recommendation effect.
To achieve the above object, the present invention discloses a supplier recommendation method, comprising:
receiving at least one supplier sample;
predicting the recommended operation corresponding to each supplier sample based on an adaptive multi-layer perception online transfer learning model;
tracking and acquiring the actual adoption effect of the supplier user on each supplier sample;
according to the comparison between the recommended operation and the actual adoption effect, the weight distribution and the nonlinear parameter vector between the user characteristics of the previous batch of the purchasing users served by the self-adaptive multi-layer perception online migration learning model and the user characteristics of the current batch are adjusted;
and re-acquiring a new supplier sample, predicting corresponding recommended operation by using the adjusted online transfer learning model, and repeating the steps until all the supplier samples meeting the supplier users are screened out.
Preferably, the prediction function of the online migration learning model based on adaptive multi-layer perception is as follows:
Figure BDA0002933553610000021
wherein, y'tAs a result of whether or not the recommended operation of the recommendation system,
Figure BDA0002933553610000022
as a projection function, α1,tAnd alpha2,tThe user characteristic weights of the previous batch and the current batch respectively correspond to weight parameters,
Figure BDA0002933553610000023
and
Figure BDA0002933553610000024
for a nonlinear projection function based on multi-layer perception,/αIs a function of the single batch prediction error rate, Θ (v, v)Φ) The final user characteristics for the previous batch,
Figure BDA0002933553610000025
the user characteristics of the current time of the batch.
Preferably, alpha1,tAnd alpha2,tThe initial values of (a) and (b) are 0.5 respectively; the update function is:
Figure BDA0002933553610000031
wherein the content of the first and second substances,
Figure BDA0002933553610000032
uφin order to be a non-linear mapping function,
Figure BDA0002933553610000033
eta is the transfer learning rate; according to the success or failure of the recommendation result, the linear part of the current feature weight of the user is updated according to the following rules:
Figure BDA0002933553610000034
Figure BDA0002933553610000035
Figure BDA0002933553610000036
wherein the content of the first and second substances,
Figure BDA0002933553610000037
user characteristics updated for the next moment, ytFor actual user operation corresponding to the actual adopted effect,
Figure BDA0002933553610000038
as a function of the loss of the hinge,
Figure BDA0002933553610000039
for learning the rate, β is the limit wtThe update speed of (2).
Preferably, in the online migration learning model based on the adaptive multi-layer perception, z is usedtAs hidden layer nodes of the nonlinear multilayer perception network, from ztTo zt+1Updating hidden layer weights in the change mode:
Figure BDA00029335536100000310
Figure BDA00029335536100000311
wherein the content of the first and second substances,
Figure BDA00029335536100000312
to learn the rate.
Preferably, in the online transfer learning model based on adaptive multi-layer perception, the method for partitioning the nonlinear mapping function into at least two linear components so that the model can continue to use the online transfer learning parameter updating mode includes:
the following nonlinear mapping function based on MLP is employed:
zt＝[zt 1,zt 2,zt 3,...,zt h]
Figure BDA00029335536100000315
wherein z ist iIs the ith hidden node, and each node is based on the ReLU function
Figure BDA00029335536100000316
H is the number of hidden nodes as an activation function; the parameter vector defining the non-linear mapping function is:
φt＝[r1 t,r2 t,...,rh t]
the updating strategy of the parameter vector updating mode obeying the PA regression algorithm is as follows:
Figure BDA0002933553610000041
Figure BDA0002933553610000042
Figure BDA0002933553610000043
wherein the content of the first and second substances,
Figure BDA0002933553610000044
is a loss function of the PA regression algorithm,
Figure BDA0002933553610000045
to learn the rate, riFor each i element of the hidden layer weight.
Preferably: lα＝ef(MC)，
Figure BDA0002933553610000046
Wherein MC is single batch prediction error rate,
Figure BDA0002933553610000047
for preference parameters, max (mc) is the current maximum prediction error rate, and min (mc) is the current minimum prediction error rate.
In order to achieve the above object, the present invention further discloses a supplier recommendation system, which includes a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the processor implements the corresponding steps of the above method when executing the computer program.
The invention has the following beneficial effects:
1. more complex non-linear recommendation problems can be handled.
2. The method has stronger robustness to misoperation of a purchasing department, and ensures that the recommendation accuracy rate is kept stable under the condition of constant demand.
3. Under the condition of changing demands, the online transfer learning model based on the adaptive multi-layer perception can quickly follow the interest change of the user, and therefore the long-term recommendation effect is improved.
The present invention will be described in further detail below with reference to the accompanying drawings.
Drawings
The accompanying drawings, which are incorporated in and constitute a part of this application, illustrate embodiments of the invention and, together with the description, serve to explain the invention and not to limit the invention. In the drawings:
FIG. 1 is a training process of a recommendation system for changes in existing user requirements.
Fig. 2 is a schematic diagram of a recommendation system architecture based on multi-level perception (MLP) adaptive online transfer learning according to an embodiment of the present invention.
FIG. 3 is a diagram illustrating a comparison between the recommendation accuracy of the recommendation system (AOTLMLP) and that of other recommendation systems according to an embodiment of the present invention.
Detailed Description
The embodiments of the invention will be described in detail below with reference to the drawings, but the invention can be implemented in many different ways as defined and covered by the claims.
Example 1
The embodiment aims to solve the long-term recommendation problem of raw material suppliers of nonferrous metallurgy enterprises, so that the built recommendation system has a better recommendation effect. For the nonferrous metallurgy purchasing department, the feature classification of the supplier is nonlinear firstly; secondly, in different time periods, the enterprise has different requirements on ore suppliers, and in the purchasing process, a decision department can make a decision (misoperation) which does not meet the requirements under the influence of the forecast value of the suppliers. According to the embodiment, a recommendation system is built by selecting an online migration learning algorithm, and nonlinear improvement and self-adaptive improvement based on multilayer perception are made on the basis of the recommendation system. The recommendation system is more suitable for the recommendations of nonferrous metallurgy suppliers.
The present embodiment is a recommendation system based on multi-layer perceptual (MLP) adaptive online migration learning, and its structure is shown in fig. 2, where AOTLMO is a recommendation algorithm module. The subscriber profile for the buyer in AOTLMO contains two components, linear and non-linear. Wherein Θ (v, v)Φ) The part of fixed features are the final user features of the previous batch and do not change along with online learning.
Figure BDA0002933553610000051
As a user characteristic of the current time of the batch,
Figure BDA0002933553610000052
user characteristics updated for the next time; the part before the comma in the user characteristic button is a linear characteristic, and the part after the comma is a nonlinear characteristic. In content-based recommendations, the same batch of provider features may be represented by a feature vector X ═ X1 x2 x3 … xn]To indicate. Wherein x istRepresenting the characteristics of the t-th supplier. In the feature vector, each feature is quantized, and for those label features that are not quantifiable, we can represent the membership of the corresponding item to the feature; the corresponding operation is the digital preprocessing of the supplier features, which is common knowledge of those skilled in the art and will not be described in detail. And the output of the recommendation system is represented by Y, wherein Y belongs to { -1,1}, 1 represents recommendation, and-1 represents non-recommendation.
The nonlinear adaptive online transfer learning recommendation algorithm adjusts the feature weight according to the comparison between the actual operation of each batch of users and the prediction result of the recommendation algorithm. First, let v and wtLinear portions of the user characteristics of the previous batch and the current tth batch, respectively. Then an item x is giventThe recommender system may predict that its label is:
Figure BDA0002933553610000053
wherein, y'tAs a result of whether or not the recommended operation of the recommendation system,
Figure BDA0002933553610000054
is a projection function. Alpha is alpha1,tAnd alpha2,tThe weight parameters are respectively corresponding to the historical characteristic weight and the current characteristic weight, and the initial values are all 0.5.
Figure BDA0002933553610000055
And
Figure BDA0002933553610000056
is a nonlinear projection function based on multi-layer perception. With constant input of samples, α1,tAnd alpha2,tIs dynamically changed, and the update function is as follows:
Figure BDA0002933553610000061
wherein the content of the first and second substances,
Figure BDA0002933553610000062
uφis a non-linear mapping function. While
Figure BDA0002933553610000063
η is the migratory learning rate. According to the success or failure of the recommendation result, the linear part of the current feature weight of the user is updated according to the following rules:
Figure BDA0002933553610000064
Figure BDA0002933553610000065
Figure BDA0002933553610000066
wherein, ytIn order for the actual user to operate,
Figure BDA0002933553610000067
as a function of the loss of the hinge,
Figure BDA0002933553610000068
to learn the rate. Again, we introduce the beta parameter to limit wtThe update speed of (2) is intended to roughly classify the nonlinear data set. After the coarse classification, the remaining classification task is completed by the non-linear part. We use ztAnd (4) de-expressing and using the node as a hidden layer node of the nonlinear multilayer perception network. Then we can get z from the PA algorithmtThe updating method of (1).
Figure BDA0002933553610000069
Figure BDA00029335536100000610
Wherein the content of the first and second substances,
Figure BDA00029335536100000611
to learn the rate. By the above steps, we obtain a product from ztTo zt+1This can be used to solve the problem of updating the hidden layer weights.
Further, in consideration of the necessity of online learning, the nonlinear mapping function is divided into a plurality of linear components so that the algorithm can continue to use the parameter updating mode of online transfer learning. Therefore we propose the following nonlinear mapping function based on MLP:
zt＝[zt 1,zt 2,zt 3,...,zt h]
Figure BDA00029335536100000614
wherein z ist iIs the ith hidden node, and each node is provided with
Figure BDA00029335536100000615
(ReLU) is used as an activation function, and h is the number of hidden nodes. The ReLU function as an activation function of the multi-layer aware network may represent any function in the case that the number of layers and the number of nodes are sufficient. Moreover, the ReLU function can be divided into two linear functions, so that the hidden layer weight can be updated according to the original linear algorithm. We define the parameter vector of the nonlinear mapping function as:
φt＝[r1 t,r2 t,...,rh t]
the updating mode of the parameter vector follows the updating strategy of the PA regression algorithm, wherein
Figure BDA0002933553610000071
Is a loss function of the PA regression algorithm,
Figure BDA0002933553610000072
to learn the rate.
Figure BDA0002933553610000073
Figure BDA0002933553610000074
Figure BDA0002933553610000075
In the above formula, riFor each i elements of the hidden layer weight, each element of the hidden layer weight can be updated according to the above update strategy.
The recommendation algorithm based on the nonlinear online transfer learning comprehensively considers the historical characteristics and the current characteristics of the previous period of time. The historical characteristics determine the stability of the recommendation accuracy rate when the demand is stable, and the current characteristics determine the following speed of the recommendation system to the demand. Under the condition that the user demand is unchanged, the prediction function is more biased to the historical characteristics, so that the synthesis is more stable. However, in the face of faster changing user demands, the recommendation system needs to make the recommendation accuracy rate after the demand changes reach the standard at a faster speed to ensure the practicability of the system. On-line migration learning, if the learning speed of the recommendation system is emphasized when the demand changes, the robustness of the recommendation system to wrong samples is reduced when the demand is stable. In order to enable the online migration recommendation algorithm to distinguish the difference between the demand change and the error sample and enable the system to achieve self-adaptation, the embodiment provides a self-adaptation online migration learning recommendation algorithm and further introduces a self-adaptation loss function described below.
In order to achieve both the speed of transfer learning and the robustness to erroneous samples, the present embodiment needs to allow the recommender system to recognize the difference between erroneous samples and the changed samples when the real demand changes. The main difference between the two methods is that the error samples are intermittently error classified in a small amount in the whole training process, and the sample change when the demand changes is continuously error classified in a large amount. To this end we introduce the following loss function:
lα＝ef(MC)
Figure BDA0002933553610000076
wherein MC is single batch prediction error rate,
Figure BDA0002933553610000077
for preference parameters, max (mc) is the current maximum prediction error rate, min (mc) is the current minimum prediction error rate, and the modified prediction function is as follows:
Figure BDA0002933553610000078
lαis a function of the predicted error rate for a single batch, and l is the predicted error rate for a single batch when the error rate is smallαThe effect of the original online transfer learning cannot be influenced, and the stability of the recommendation accuracy under the state of unchanged demand is maintained. And under the condition of large error rate of single batch, the algorithm considers that the demand change occurs, iαWill increase and speed up the adjustment of the procurement sector demand characteristics. lαThe introduction of the adaptive loss function enables online transfer learning to distinguish between erroneous samples and changes in demand in the training sequence.
In summary, the parameter adjustment of the present embodiment includes the following aspects:
in the recommendation algorithm based on the adaptive online transfer learning, parameters needing to be set comprise a loss function limiting parameter beta, a transfer learning rate eta and a preference parameter
Figure BDA0002933553610000081
First, for the selection of β, a more complex nonlinear data set needs to reduce β to allow more classification tasks to be performed in the nonlinear part. In setting η, project information complexity (number of features, number of pieces of data) needs to be considered. And adjusting eta when the demand is stable to maintain the recommendation accuracy within an interval to ensure the robustness of the recommendation system, and enabling the recommendation accuracy to be recovered to a normal level after the demand changes. Too small η ignores the effect of the historical feature weights, making the recommendation system less robust, while too large η causes the system to diverge.
Figure BDA0002933553610000082
The stability and migration capabilities of the recommendation system are determined. If the recommendation system has higher requirements on the stability of the recommendation accuracy rate, the requirements are reduced
Figure BDA0002933553610000083
The value of (c). Conversely, if the recommendation system emphasizes the following speed to the user's needs, then it needs to be increased
Figure BDA0002933553610000084
The value of (c). In the classification problem
Figure BDA0002933553610000085
Possibly between 0 and 1.
Therefore, the embodiment discloses a specific recommendation algorithm comprising the following steps:
inputting: feature set { x) of all items in the batch1 … xTH, and the previous stage end user characteristics V, VΦThe error rate MC of the previous batch prediction, preference parameter
Figure BDA0002933553610000086
Loss function limiting parameter β, transfer learning rate η, batch size T.
Initialization: w is a1＝0,α1,t＝α2,t＝0.5,Φt＝[0.5,0.5,...,0.5]
Step 1: receiving a sample xt；
Step 2: calculating ztAnd predicting operation y 'corresponding to the sample't；
zt＝[zt 1,zt 2,zt 3,...,zt h]
Figure BDA0002933553610000087
Receiving the actual operation y of the user on the samplet；
Number of mispredictions if classification is wrong: m + 1; otherwise, returning to the second step;
and step 3: calculate the new:
Figure BDA0002933553610000091
and to wt，ztUpdating:
Figure BDA0002933553610000092
Figure BDA0002933553610000093
and 4, step 4: according to ztIs to the non-linear parameter vector phitUpdating:
Figure BDA0002933553610000094
and 5: repeating the steps until T, and calculating the error prediction rate of the batch:
Figure BDA0002933553610000095
in order to verify the effectiveness of the adaptive online transfer learning supplier recommendation algorithm of the embodiment, purchasing decision and purchasing data of a certain purchasing department of a zinc smelting plant in a period of time and a characteristic set of a supplier are used as cases. For the characteristics of the suppliers, such as economic factors, environmental factors, social factors and the like, the purchasing department can select a group of suppliers which are suitable for the production requirements in a current period of time, so that the enterprise benefit is maximized. For zinc smelting suppliers, 11 characteristics including price level, zinc content, lead content, silicon content, accuracy of zinc forecast value, accuracy of lead forecast value, accuracy of silicon forecast value, accuracy of arrival date, standard reaching rate of supply quantity, natural geographic environment and enterprise development prospect are considered. By adjusting the parameters, we consider that when β is 0.9, η is 0.05,
Figure BDA0002933553610000096
the recommendation system works best. In this case we compared the original online migration algorithm to the non-linearityThe recommendation effect of the recommendation system constructed by the adaptive online migration algorithm and the PAMO nonlinear online algorithm is shown in fig. 3.
Compared with the traditional recommendation system based on OTL, the recommendation system based on AOTLMO can process the classification problem of the nonlinear data set, thereby better completing the recommendation. And secondly, compared with a recommendation system based on a common nonlinear online algorithm (PAMO), the AOTLMO recommendation system can relieve the fluctuation of recommendation accuracy rate caused by misoperation. Compared with a recommendation system formed by an online migration algorithm (OTLMO) based on MLP, the AOTLMO recommendation system can follow the demand more quickly after the demand changes and ensure quick reply of the recommendation accuracy. After the demand changes, the recommendation accuracy of the AOTLMO algorithm can be recovered to more than 80% after about 20 batches, while the recommendation accuracy of the OTLMO algorithm is always at a lower level, and the accuracy begins to recover after 50 batches. The results show that the AOTLMO recommendation system not only can keep stable and high recommendation accuracy under the condition of stable demand, but also can adapt to user characteristics more quickly after the demand changes, and reflects the consideration of historical data and online adjustment capability.
Example 2
The embodiment discloses a supplier recommendation system, which comprises a memory, a processor and a computer program stored on the memory and capable of running on the processor, wherein the processor executes the computer program to realize the steps corresponding to the method in the embodiment.
To sum up, the supplier recommendation method and system disclosed by the embodiment of the invention have the following beneficial effects:
1. more complex non-linear recommendation problems can be handled.
2. The method has stronger robustness to misoperation of a purchasing department, and ensures that the recommendation accuracy rate is kept stable under the condition of constant demand.
3. Under the condition of changing demands, the online transfer learning model based on the adaptive multi-layer perception can quickly follow the interest change of the user, and therefore the long-term recommendation effect is improved.
The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention, and various modifications and changes may be made by those skilled in the art. Any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims (7)

1. A supplier recommendation method, comprising:
receiving at least one supplier sample;
predicting the recommended operation corresponding to each supplier sample based on an adaptive multi-layer perception online transfer learning model;
tracking and acquiring the actual adoption effect of the supplier user on each supplier sample;
according to the comparison between the recommended operation and the actual adoption effect, the weight distribution and the nonlinear parameter vector between the user characteristics of the previous batch of the purchasing users served by the self-adaptive multi-layer perception online migration learning model and the user characteristics of the current batch are adjusted;
and re-acquiring a new supplier sample, predicting corresponding recommended operation by using the adjusted online transfer learning model, and repeating the steps until all the supplier samples meeting the supplier users are screened out.
2. The supplier recommendation method according to claim 1, wherein the prediction function of the online migration learning model based on the adaptive multi-layer perception is as follows:
Figure FDA0002933553600000011
wherein, y'iAs a result of whether or not the recommended operation of the recommendation system,
Figure FDA0002933553600000012
as a projection function, α1,tAnd alpha2,tRespectively the user characteristic weight of the previous batch and the user characteristic of the current batchThe weight parameter corresponding to the characteristic weight is obtained,
Figure FDA0002933553600000013
and
Figure FDA0002933553600000014
for a nonlinear projection function based on multi-layer perception,/αIs a function of the single batch prediction error rate, Θ (v, v)Φ) The final user characteristics for the previous batch,
Figure FDA0002933553600000015
the part before the comma in the user characteristic mark is a linear characteristic, and the part after the comma is a nonlinear characteristic.
3. The supplier recommendation method of claim 2, wherein α is1,tAnd alpha2,tThe initial values of (a) and (b) are 0.5 respectively; the update function is:
Figure FDA0002933553600000016
wherein the content of the first and second substances,
Figure FDA0002933553600000017
u∈Rd，uφin order to be a non-linear mapping function,
Figure FDA00029335536000000110
eta is the transfer learning rate; according to the success or failure of the recommendation result, the linear part of the current feature weight of the user is updated according to the following rules:
Figure FDA0002933553600000018
Figure FDA0002933553600000019
Figure FDA00029335536000000111
wherein the content of the first and second substances,
Figure FDA0002933553600000021
user characteristics updated for the next moment, ytFor actual user operation corresponding to the actual adopted effect,
Figure FDA00029335536000000210
as a function of the loss of the hinge,
Figure FDA0002933553600000022
for learning the rate, β is the limit wtThe update speed of (2).
4. The vendor recommendation method of claim 3, further comprising:
in the online migration learning model based on the adaptive multi-layer perception, z is usedtAs hidden layer nodes of the nonlinear multilayer perception network, from ztTo zt+1Updating hidden layer weights in the change mode:
Figure FDA0002933553600000023
Figure FDA0002933553600000024
wherein the content of the first and second substances,
Figure FDA0002933553600000025
to learn the rate.
5. The vendor recommendation method of claim 4, further comprising:
in the online transfer learning model based on the adaptive multi-layer perception, a parameter updating mode for dividing a nonlinear mapping function into at least two linear components so that the model can continuously use online transfer learning comprises the following steps:
the following nonlinear mapping function based on MLP is employed:
zt＝[zt 1,zt 2,zt 3,...,zt h]
Figure FDA00029335536000000212
wherein z ist iIs the ith hidden node, and each node is based on the ReLU function
Figure FDA00029335536000000213
H is the number of hidden nodes as an activation function; the parameter vector defining the non-linear mapping function is:
φt＝[r1 t,r2 t,...,rh t]
the updating strategy of the parameter vector updating mode obeying the PA regression algorithm is as follows:
Figure FDA0002933553600000026
Figure FDA0002933553600000027
Figure FDA00029335536000000214
wherein the content of the first and second substances,
Figure FDA00029335536000000211
is a loss function of the PA regression algorithm,
Figure FDA0002933553600000028
to learn the rate, riFor each i element of the hidden layer weight.
6. The supplier recommendation method according to any one of claims 2 to 5, characterized in that:
lα＝ef(MC)
Figure FDA0002933553600000029
wherein MC is single batch prediction error rate,
Figure FDA0002933553600000031
for preference parameters, max (mc) is the current maximum prediction error rate, and min (mc) is the current minimum prediction error rate.
7. A supplier recommendation system comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor when executing the computer program implements the steps of the method of any of the preceding claims 1 to 6.
CN202110156145.0A 2021-02-04 2021-02-04 Vendor recommendation method and system Active CN112927037B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110156145.0A CN112927037B (en) 2021-02-04 2021-02-04 Vendor recommendation method and system

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110156145.0A CN112927037B (en) 2021-02-04 2021-02-04 Vendor recommendation method and system

Publications (2)

Publication Number Publication Date
CN112927037A true CN112927037A (en) 2021-06-08
CN112927037B CN112927037B (en) 2024-02-06

Family

ID=76170393

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110156145.0A Active CN112927037B (en) 2021-02-04 2021-02-04 Vendor recommendation method and system

Country Status (1)

Country Link
CN (1) CN112927037B (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115358781A (en) * 2022-08-22 2022-11-18 陕西师范大学 Crowd sensing noise monitoring task recommendation method based on limited rational decision model

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20040243527A1 (en) * 2003-05-28 2004-12-02 Gross John N. Method of testing online recommender system
US20170132509A1 (en) * 2015-11-06 2017-05-11 Adobe Systems Incorporated Item recommendations via deep collaborative filtering
CN107016122A (en) * 2017-04-26 2017-08-04 天津大学 Knowledge recommendation method based on time-shift
CN109190053A (en) * 2018-07-04 2019-01-11 南京邮电大学 One kind being based on point of interest importance and the authoritative point of interest recommended method of user

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20040243527A1 (en) * 2003-05-28 2004-12-02 Gross John N. Method of testing online recommender system
US20170132509A1 (en) * 2015-11-06 2017-05-11 Adobe Systems Incorporated Item recommendations via deep collaborative filtering
CN107016122A (en) * 2017-04-26 2017-08-04 天津大学 Knowledge recommendation method based on time-shift
CN109190053A (en) * 2018-07-04 2019-01-11 南京邮电大学 One kind being based on point of interest importance and the authoritative point of interest recommended method of user

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
樊海玮等: "基于MLP 改进型深度神经网络学习资源推荐算法", 《计算机应用研究》, vol. 37, no. 9, pages 2629 - 2633 *

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115358781A (en) * 2022-08-22 2022-11-18 陕西师范大学 Crowd sensing noise monitoring task recommendation method based on limited rational decision model
CN115358781B (en) * 2022-08-22 2023-04-07 陕西师范大学 Crowd sensing noise monitoring task recommendation method based on limited rational decision model

Also Published As

Publication number Publication date
CN112927037B (en) 2024-02-06

Similar Documents

Publication Publication Date Title
Hubbs et al. A deep reinforcement learning approach for chemical production scheduling
Ji et al. An Application of a Three‐Stage XGBoost‐Based Model to Sales Forecasting of a Cross‐Border E‐Commerce Enterprise
Huang et al. A hybrid model for stock market forecasting and portfolio selection based on ARX, grey system and RS theories
Xia et al. A seasonal discrete grey forecasting model for fashion retailing
Torabi et al. ABC inventory classification in the presence of both quantitative and qualitative criteria
Krasheninnikova et al. Reinforcement learning for pricing strategy optimization in the insurance industry
Wang et al. A trend fixed on firstly and seasonal adjustment model combined with the ε-SVR for short-term forecasting of electricity demand
Hadavandi et al. An improved sales forecasting approach by the integration of genetic fuzzy systems and data clustering: Case study of printed circuit board
CN114219169A (en) Script banner supply chain sales and inventory prediction algorithm model and application system
CN109727078B (en) Sales prediction correction method based on commodity category tree
CN108960474B (en) Warehouse site selection method for online retailer in warehouse network considering cross
Arredondo et al. Learning and adaptation of a policy for dynamic order acceptance in make-to-order manufacturing
US20190378061A1 (en) System for modeling the performance of fulfilment machines
Chang et al. A hybrid system by evolving case-based reasoning with genetic algorithm in wholesaler's returning book forecasting
Badejo et al. Integrating tactical planning, operational planning and scheduling using data-driven feasibility analysis
Gandhmal et al. Wrapper-enabled feature selection and CPLM-based NARX model for stock market prediction
Rahman et al. Supplier selection using fuzzy-topsis method: A case study in a cement industry
Wang et al. A neuro-fuzzy based forecasting approach for rush order control applications
CN112927037A (en) Supplier recommendation method and system
KR102590158B1 (en) Machine learning based demand predicting system and demand predicting method
Garnier Concurrent neural network: a model of competition between times series
US20240168467A1 (en) Computer-Implemented Methods Referring to an Industrial Process for Manufacturing a Product and System for Performing Said Methods
Li et al. Zinc ore supplier evaluation and recommendation method based on nonlinear adaptive online transfer learning.
Liu Using geometric programming to profit maximization with interval coefficients and quantity discount
CN114612005A (en) Glass factory digital intelligent system based on software as a service (SaaS) implementation

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
GR01 Patent grant
GR01 Patent grant