CN110264364B - Recommendation method for investor - Google Patents
Recommendation method for investor Download PDFInfo
- Publication number
- CN110264364B CN110264364B CN201910359691.7A CN201910359691A CN110264364B CN 110264364 B CN110264364 B CN 110264364B CN 201910359691 A CN201910359691 A CN 201910359691A CN 110264364 B CN110264364 B CN 110264364B
- Authority
- CN
- China
- Prior art keywords
- node
- investor
- label
- nodes
- investment
- 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.)
- Active
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
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- 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
- G06Q40/00—Finance; Insurance; Tax strategies; Processing of corporate or income taxes
- G06Q40/06—Asset management; Financial planning or analysis
Landscapes
- Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Human Resources & Organizations (AREA)
- Strategic Management (AREA)
- Development Economics (AREA)
- Economics (AREA)
- Marketing (AREA)
- Physics & Mathematics (AREA)
- Operations Research (AREA)
- Finance (AREA)
- Entrepreneurship & Innovation (AREA)
- Accounting & Taxation (AREA)
- Theoretical Computer Science (AREA)
- Game Theory and Decision Science (AREA)
- General Business, Economics & Management (AREA)
- General Physics & Mathematics (AREA)
- Technology Law (AREA)
- Quality & Reliability (AREA)
- Tourism & Hospitality (AREA)
- Financial Or Insurance-Related Operations Such As Payment And Settlement (AREA)
Abstract
The invention discloses a sponsor recommendation method which is used for improving the accuracy of recommending sponsors for an initial company. In the recommendation method, historical investment event information and label information are obtained from a historical investment database; then generating an investment network according to the historical investment event information and the label information, wherein the investment network comprises: m investor nodes, N investor nodes, P label nodes and the like; performing, for a plurality of iterations, a diffusion process comprising: distributing the score obtained by the first type node in the investment network in the current diffusion step to a first neighbor node connected with the first type node, wherein the first neighbor node uses the obtained score in the next diffusion step; when the executed diffusion step reaches the optimal diffusion step number, ending the diffusion process; and sequencing the M investor nodes from large to small according to the final scores respectively possessed, and recommending at least one investor node to a target company to be recommended according to a sequencing result.
Description
Technical Field
The invention relates to the technical field of computers, in particular to a recommendation method for an investor.
Background
The internet is developed rapidly at present, users enter an information explosion era, and the presentation of massive information makes people difficult to find useful information and make correct decisions.
The recommendation system is an effective tool for solving the information overload problem. In the field of goods or services such as movies, books, music and the like, the recommendation system plays an important role, provides accurate personalized recommendation for users, and in the field of financial investment, the recommendation problem is rarely related.
Compared with the purchasing or using data of the goods or the services, the investment relation in the financial investment field is sparser, and the recommendation algorithm provided by the prior art is only suitable for the recommendation of the goods or the services, but not suitable for the investment recommendation in the financial investment field, so that the recommendation model and the recommendation algorithm provided by the prior art cannot achieve higher recommendation accuracy.
For the initial companies seeking investments, especially for the most of the companies that never acquire them, it is often necessary to spend a lot of time and effort to find the investors who may be interested in their field due to the lack of experience. The implementation of recommendations for such companies is called the cold start problem in recommendation system research, i.e. in case that the historical invested situation of the target object is unknown, it is difficult to apply the traditional recommendation method to implement personalized recommendations for it.
Therefore, a sponsor recommendation method for the original company is urgently needed to overcome the problem of low recommendation accuracy of the traditional recommendation algorithm.
Disclosure of Invention
The invention aims to provide a sponsor recommending method, which is used for improving the accuracy of recommending sponsors for an initial company.
In order to achieve the purpose, the invention adopts the following technical scheme:
the invention provides a recommendation method of an investor, which comprises the following steps:
obtaining historical investment event information and label information from a historical investment database, wherein the historical investment event information is used for indicating a historical investment relation between an xth investor in M investors and an ith investor in N investors, the label information comprises labels corresponding to the N investors respectively, each investor in the N investors comprises at least one label, the M, the N, the x and the i represent positive integers, the x is smaller than or equal to the M, and the i is smaller than or equal to the N;
generating an investment network according to the historical investment event information and the label information, wherein the investment network comprises: m investor nodes, N investor nodes, P label nodes, a first type connecting edge between the investor node and the investor node, and a second type connecting edge between the investor node and the label node; the M investor nodes correspond to the M investors one by one, the N investor nodes correspond to the N invested parties one by one, the P label nodes correspond to all different labels corresponding to the N invested parties one by one, the first type connecting edge indicates that an investment relation is established between the x-th investor node and the i-th investor node, and the second type connecting edge indicates that an investment relation is established between the x-th investor node and the label node corresponding to the label included in the i-th investor node;
performing, for a plurality of iterations, a diffusion process comprising: distributing the score obtained by a first type node in the investment network in the current diffusion step to a first neighbor node connected with the first type node, wherein the first neighbor node uses the obtained score in the next diffusion step, label nodes corresponding to all labels included by a target company to be recommended in the investment network have initial scores in the diffusion process, and the M investor nodes, the N investor nodes and the P label nodes in the investment network do not have initial scores in the diffusion process except the label nodes corresponding to all labels included by the target company; wherein, when the first type node is the label node, the first neighbor node comprises: the sponsor node is connected with the label node through the second type connecting edge; when the first type node is the sponsor node, the first neighbor node comprises: the label node is connected with the sponsor node through the second type connecting edge, and the sponsor node is connected with the sponsor node through the first type connecting edge; when the first type node is the sponsor node and the label node, the first neighbor node comprises: a sponsor node connected with the sponsor node through the first type connecting edge and a sponsor node connected with the label node through the second type connecting edge;
determining whether the executed diffusion step of the diffusion process has reached a predetermined optimal number of diffusion steps, and ending the diffusion process when the executed diffusion step has reached the optimal number of diffusion steps;
and obtaining final scores which are respectively possessed by the M investor nodes in the investment network when the diffusion process is finished, sequencing the M investor nodes from large to small according to the final scores which are respectively possessed, and recommending at least one investor node to the target company according to a sequencing result.
After the technical scheme is adopted, the technical scheme provided by the invention has the following advantages:
in an embodiment of the present application, an investment network is generated, the investment network comprising: m investor nodes, N investor nodes, P label nodes, a first type connecting edge between the investor node and the investor node, and a second type connecting edge between the investor node and the label node. In the embodiment of the invention, the diffusion process can be iteratively executed for a plurality of times, so that the initial values of the label nodes in the diffusion process can be continuously diffused in the investment network, because the connecting edges of the three parts of nodes in the investment network are connected according to the real contact, and the investor node is connected with the invested node (namely the invested company, called the company for short) and the invested label node. Therefore, the scores are diffused along the connecting edges, and the nodes with indirect connection are all diffused. For example, the initial score is on the label nodes of the target company, in the first diffusion, the investor nodes investing the labels receive a certain score, and the investors who invest the labels more receive more scores. Then the investor nodes spread the scores to the company nodes and the label nodes, the company nodes obtaining the scores are invested companies of investors who invest the labels contained in the target company, and the label nodes obtaining the scores are invested labels of investors who invest the labels contained in the target company. That is, the sponsor nodes, the company nodes, and the tag nodes that directly relate to the tags receive the scores in succession, and the scores are then collected in a set of sponsor nodes. Therefore, the weak relationship is continuously strengthened through diffusion and finally reflected in the score of the node of the investor, the larger the score is, the stronger the relationship between the investor and the target company is, the higher the possibility of generating investment relationship between the investor and the target company is, namely, the investor with high score is recommended to the target company. Therefore, the accuracy rate of recommending the investor for the initial company can be improved.
Drawings
FIG. 1 is a block diagram illustrating a process of a method for recommending an investor in accordance with an embodiment of the present invention;
FIG. 2 is a schematic diagram of the structure of an investment network provided by an embodiment of the present invention;
FIG. 3 is a schematic diagram illustrating connection relationships between nodes in an investment network according to an embodiment of the present invention;
FIG. 4a is a schematic diagram of initial point setting in an investment network according to an embodiment of the present invention;
FIG. 4b is a schematic diagram of a first step spread within an investment network provided by an embodiment of the present invention;
FIG. 4c is a schematic diagram of a second step of flooding within the investment network provided by an embodiment of the present invention;
fig. 4d is a schematic diagram of a third step of flooding within the investment network provided by an embodiment of the present invention.
Detailed Description
The embodiment of the invention provides a sponsor recommendation method, which is used for improving the accuracy of sponsor recommendation for an initial company.
The following are detailed below.
The embodiment of the recommendation method of the investor can be applied to recommending the investor with high matching degree with the initial company to the initial company. In the subsequent embodiments of the present invention, the investor node may be referred to as investor, the invested node may be referred to as invested company, and the target invested node to be recommended may be referred to as target company. In the embodiment of the invention, through the diffusion process described later, the strong and weak relationship between the investor and the target company and the possibility of generating the investment relationship can be reflected through the score obtained by the investor, and the investor with high score is finally recommended to the target company. Therefore, the accuracy rate of recommending the investor for the initial company can be improved.
Referring to fig. 1, the method for recommending an investor according to the present invention may include the following steps:
10. historical investment event information and label information are obtained from a historical investment database, and the historical investment event information is used for indicating the historical investment relation between the xth investor in the M investors and the ith investor in the N investors. The label information comprises labels corresponding to N invested parties respectively, each invested party in the N invested parties comprises at least one label, M, N, x and i represent positive integers, x is smaller than or equal to M, and i is smaller than or equal to N.
In the embodiment of the invention, a historical investment database is configured in advance, and historical investment event information and label information are stored in the historical investment database, so that the historical investment event information and the label information can be acquired from the historical investment database. The historical investment event information may also be referred to as historical investment relationship data in embodiments of the present invention to indicate a historical investment relationship between the xth of the M investors and the ith of the N investors. The values of M and N may be determined according to a specific scenario. The label information is used to indicate labels corresponding to the N invested parties respectively, each invested party includes at least one label, and the content component of the label is not limited here. In addition, the M, N, x and i in the embodiment of the invention both represent positive integers, x is less than or equal to M, and i is less than or equal to N.
In some embodiments of the invention, the historical investment event information comprises: the identifier of the x-th investor, the identifier of the i-th invested party and investment record data.
For example, historical investment relationship data is obtained over a period of time, including all investment event information and label information for the invested company. The investment event information includes information of investors and invested companies, investment time, investment amount, share ratio and the like related to the event, for example, the format can be investor ID, invested company ID, investment time, investment amount, share ratio and other investment information, the information can be obtained from investment data providers, for example, the information can be obtained from investment data providers such as IT orange, CVSource investment data, inventory data and the like, and some investors or companies can also disclose recent investment conditions.
In the embodiment of the present invention, the tag is composed of some keywords, for example, the format of the tag may be a company ID: tag 1, tag 2, … … tag n. The labels reflect the industry field and the main business of the company, the labels of the company can be obtained by searching and inquiring from channels such as the data service providers, a label library can be built by self, and key information is extracted according to the contents such as basic introduction, product information, commercial behaviors, news dynamics, public evaluation and the like of the company to set the labels for the company.
20. Generating an investment network according to the historical investment event information and the label information, wherein the investment network comprises: m investor nodes, N investor nodes, P label nodes, a first type connecting edge between the investor node and the investor node, and a second type connecting edge between the investor node and the label node.
The first type connecting edge indicates that an investment relation is established between the x-th investor node and the ith investor node, and the second type connecting edge indicates that an investment relation is established between the x-th investor node and the label node corresponding to the label included by the ith investor node.
In some embodiments of the present invention, after the historical investment event information and the tag information are obtained, information of an investor, an invested company, a tag, and the like may be extracted according to data content recorded in the historical investment event information and the tag information, and each investor corresponds to one node in the network, each invested company corresponds to one node in the network, and each tag corresponds to one node in the network.
As shown in fig. 2, an example architecture of an investment network. If M investor nodes, N investor nodes and P label nodes are determined, an investment network is generated by the M investor nodes, the N investor nodes and the P label nodes, and the investment network comprises the three nodes and also comprises the connection relation among the nodes. Specifically, M investor nodes are located in the middle column of the investment network, and N invested nodes and P label nodes are located on two sides of the M investor nodes respectively. For example, N sponsored nodes are located to the left of M sponsor nodes, and P tag nodes are located to the right of M sponsor nodes. The investment network comprises the three nodes, a first type connecting edge between an investor node and a invested node and a second type connecting edge between the investor node and a label node, wherein the connecting edges are established between the investor node and the invested node, and the connecting edges are established between the investor node and the label node.
In the embodiment of the present invention, two types of connecting edges in the embodiment of the present invention are schematically illustrated by dotted lines in fig. 2. The first type connecting edge indicates that an investment relation is established between the x-th investor node and the i-th investor node, and the second type connecting edge indicates that an investment relation is established between the x-th investor node and a label node corresponding to a label included by the i-th investor node. For example, the first type connecting edge refers to an investment relationship between a supplier node and a supplier node, if the investment relationship exists between the supplier node and the supplier node, the first type connecting edge is established between the two nodes of the investment network, and if the investment relationship does not exist between the supplier node and the supplier node, the first type connecting edge is not established between the two nodes of the investment network. Similarly, the second type connecting edge indicates that an investment relationship is established between the investor node and the label node, if the investor node invests a company containing a certain label, the second type connecting edge is established between the two nodes of the investment network, and if the investor node does not invest the company containing the certain label, the second type connecting edge is not established between the two nodes of the investment network.
In some embodiments of the present application, the historical investment event information comprises: the identifier of the xth investor, the identifier of the ith invested party and investment record data;
the ith invested party node comprises at least one label used for reflecting the industry field and the main operation business of the ith invested party.
Wherein the label reflects the industry field and the main operation business of the corresponding invested party node. For example, a company label is: tourism, tourism integrated services, hotels, visa services, travel ticketing and the like.
In some embodiments of the present application, after the step 20 of generating the investment network according to the historical investment event information and the tag information, the method provided by the embodiments of the present application further includes the following steps:
determining the weight of the first type connecting edge according to investment record data between the investor node and the invested party node;
and determining the weight of the second type connecting edge according to the investment relation between the investor node and the invested party node, the affiliated relation between the invested party node and the label and the investment record data between the investor node and the invested party node.
The investment record data can include investment amount, share ratio and other investment information, the weight of the first type of connecting edge is determined according to the investment record data between the investor node and the investor node, and similarly, the weight of the second type of connecting edge can be established. When the first-type continuous edge and the second-type continuous edge both have the weight, the score in subsequent diffusion based on the investment network needs to be diffused according to the weight, which will be described in detail later.
In the embodiment of the invention, three data sets can be obtained: investor set I, invested company set C and label set T, and two data relationships: investments relationships between investors-companies and affiliations between company-tags. In empirical analysis, the investor has obvious investment preference for specific fields and labels, so the embodiment of the invention takes the invested company as a bridge to link the investor and the labels, thereby utilizing the preference information in recommendation.
Fig. 3 is a schematic diagram of a connection relationship between nodes in an investment network according to an embodiment of the present invention. Constructing a company-supplier-label three-part graph G (C, I, T) according to the acquired data, wherein the construction rule is as follows: each investor, invested company and label are represented by nodes, the investment relation is represented by connecting edges, the company and the label are arranged at two sides of the three-part graph, the company and the label are connected with the corresponding investor, but the connecting edges are not established between the company and the label, if the investor I x Invested company C i Then at I x And C i Establishing a connecting edge; if Simultaneous company C i Including a tag T a And T b Then at I x And T a 、T b The connection edge is established between the nodes in the set, namely, the connection edge is not formed between companies, the connection edge is not formed between labels, and the connection edge is not formed between investors. The weight of the connecting side between the company and the investor is quantified by actual numerical values such as investment amount, share ratio and the like, and the weight of the connecting side between the investor and the label is determined by the number of companies comprising the label invested by the investor, the investment amount of each investment, the share ratio and the like.
For example, if the investor group comprises 4 investors, I1, I2, I3 and I4 respectively, the invested company group comprises 6 companies C1 to C6, and the tag group comprises 4 tags, T1 to T4; i1 invests C2 and C4, I2 invests C3 and C5, I3 invests C4, C5 and C6, and I4 invests C5; the label of C1 is T1, T2, the label of C2 is T1, T2, T4, the label of C3 is T1, T2, T4, the label of C4 is T3, T4, the label of C5 is T4, and the label of C6 is T4, where the embodiment of the present invention uses a simple weighting rule as an example, the edge weights between the investor and the company are all regarded as 1, the edge weights between the investor and the label are set as the number of companies including the label in the investor investment, and a company-investor-label three-part diagram as shown in fig. 3 can be established according to the investment data, and the numbers on the edges in the diagram represent the weight size. Since the target company to be recommended does not have any invested experience, it will not be connected to any investor in the constructed three-part diagram, such as company C1 in fig. 3.
30. Performing, for a plurality of iterations, a diffusion process comprising: and distributing the score obtained by the first type node in the investment network in the current diffusion step to a first neighbor node connected with the first type node, wherein the first neighbor node uses the obtained score in the next diffusion step.
The method comprises the steps that label nodes corresponding to all labels included by a target company to be recommended in an investment network have initial scores in the diffusion process, and M investor nodes, N investor nodes and P label nodes in the investment network do not have initial scores in the diffusion process except the label nodes corresponding to all the labels included by the target company; wherein, when the first type node is a label node, the first neighbor node comprises: the investor node is connected with the label node through a second type connecting edge; when the first type node is a sponsor node, the first neighbor node comprises: the label node is connected with the investor node through the second type connecting edge, and the investor node is connected with the investor node through the first type connecting edge; when the first type node is a sponsor node and a label node, the first neighbor node comprises: the system comprises a sponsor node connected with a sponsor node through a first type connecting edge and a sponsor node connected with a label node through a second type connecting edge.
In the embodiment of the present invention, after the investment network is generated through step 20, the following diffusion processes may be performed in a plurality of iterations, where each diffusion process includes the following two processes: and distributing the score obtained by the first type node in the investment network in the current diffusion step to a first neighbor node connected with the first type node, wherein the first neighbor node uses the obtained score in the next diffusion step. Namely, after the first neighbor node obtains the score, the first neighbor node is used as the first type node to perform diffusion again, namely, the score obtained by the first type node in the investment network in the current diffusion step is distributed to the first neighbor node connected with the first type node.
It should be noted that, in this embodiment of the present invention, when the first type node is a sponsor node and a tag node, the first neighbor node includes: the system comprises a sponsor node connected with a sponsor node through a first type connecting edge and a sponsor node connected with a label node through a second type connecting edge. Both sides can be diffused to the middle through connecting the sides, the investor can be connected with the left end or the right end, or can be connected with the left end and the right end at the same time, the left side can be connected with the left side to obtain the score diffused from the left side, and the right side can be connected with the right side to obtain the score diffused from the right side.
In this embodiment of the present invention, when both the first-type continuous edge and the second-type continuous edge have a weight, the assigning, in step 30, the score obtained by the first-type node in the investment network in the current diffusion step to the first neighbor node connected to the first-type node includes:
and averagely distributing the score obtained by the first type node in the investment network in the current diffusion step to the first neighbor node according to the weight of the first type connecting edge and the weight of the second type connecting edge.
Under the condition that the first type continuous edge and the second type continuous edge both have weights, scores obtained in the current diffusion step of the first type node in the investment network need to be diffused according to the weights when the diffusion is performed on the basis of the investment network, and the scores are averagely distributed to the first neighbor nodes according to the weights of the first type continuous edge and the second type continuous edge.
In some embodiments of the present invention, the diffusion process in step 30 may specifically include the following steps:
step a: in the initial step before the beginning of diffusion, setting initial scores for label nodes corresponding to all labels included by a target company to be recommended in an investment network, and setting the scores to be 0 for other label nodes except the label nodes corresponding to all labels included by the target company in M investor nodes, N investor nodes and P label nodes in the investment network;
step b: when the first type node is a label node, averagely distributing the initial score of the label node to the investor node through a second type connecting edge according to the weight; after the step b is executed, sequentially and iteratively executing the following steps c and d until the optimal diffusion step number is reached, and stopping iteration;
step c: when the first type node is the investor node, the value obtained by the investor node through the step b is averagely distributed to the label node and the investor node through the second type connecting edge and the first type connecting edge according to the weight;
step d: and when the first type nodes are the invested node and the label node, the value obtained by the invested node through the step c and the value obtained by the label node through the step c are averagely distributed to the investor node through the first type connecting edge and the second type connecting edge according to the weight, the investor node superposes the values obtained through the first type connecting edge and the second type connecting edge, and the investor node triggers the step c to be executed through the value obtained by the step d.
In the foregoing steps c and d, it is necessary to perform iteration according to a plurality of times. In order to recommend a proper investor to a target company, on an investment network (a company-investor-label three-part graph), the embodiment of the invention endows each label in the label set with a certain initial score, for example, a label connected with the target company obtains an initial score of 1 unit, and the initial scores of other labels are set to be 0. These scores will then be diffused by connecting edges in the three-part graph. In each step of diffusion, each node (investor, company or label) containing the score distributes the score to the adjacent nodes according to the weight average, and each node adds the obtained scores to obtain the new score of the node. Through the odd diffusion, the score will fall into the sponsor pool. The score after each step of diffusion is calculated as follows:
it should be noted that the diffusion process may be bidirectional, in each diffusion step, a node containing a score distributes the score to a neighboring node, after the score is distributed, the score of the node is cleared, and then all nodes add the newly accepted scores (distributed from other nodes) to set the new score of the node.
For example, in the initial stage, the initial score is in the label node set, so in the first step of diffusion, the nodes containing the score are only the label nodes contained in the target company, and the label nodes distribute the scores of the nodes to the connected sponsor nodes; then, carrying out a second step of diffusion, wherein the nodes containing the scores are a plurality of investor nodes, and the investor nodes distribute the scores of the nodes to the connected company nodes and the connected label nodes; thirdly, diffusing, wherein the nodes containing the scores are a plurality of company nodes and label nodes, the company nodes distribute the scores of the company nodes to the connected investor nodes, and the label nodes also distribute the scores of the company nodes to the connected investor nodes; the fourth step is to spread, this time the points containing the scores are the sponsor nodes, which distribute their scores to the connected company nodes and label nodes, and so on. Therefore, the point value is distributed from the investor node (the middle node set) to both sides, because the investor node is connected with the company node and the label node. When the scores are diffused from the company nodes or the label nodes, the scores are diffused in a single direction, because the company nodes are only connected with the investor nodes, and the label nodes are only connected with the investor nodes.
For example, f (I) y ) Is a node I y Score before diffusion, f' (I) by this step y ) For the score after diffusion, A is the investor-company adjacency matrix, if investor I y Invested company C x Then A (I) y ,C x )＝W yx ，W yx Quantizing according to actual numerical values such as investment amount, share ratio and the like, and otherwise, the actual numerical values are 0; a' is the sponsor-label adjacency matrix if sponsor I y Invested over-containing tag T z Company of (1), A' (I) y ,T z )＝W yz ，W yz The number of the label companies which are invested by the investor, the investment amount of each investment, the share ratio and other numerical values are determined, and if not, the number is 0; k is a radical of Iy→C Indicating the sponsor I y Sum of weights, k, of edges connecting all companies Iy→T Supplier I y Sum of weights, k, of edges connecting all labels Cx And k Tz Respectively represent company C x And a label T z The sum of the weights of all investors connected; lambda is an adjusting parameter and is used for adjusting the value proportion of the spreading of the investor node to the left and right sides, and the lambda can be trained from historical data to obtain an optimal value.
Fig. 4a is a schematic diagram of initial score setting in the investment network according to the embodiment of the present invention, fig. 4b is a schematic diagram of a first step diffusion in the investment network according to the embodiment of the present invention, fig. 4c is a schematic diagram of a second step diffusion in the investment network according to the embodiment of the present invention, and fig. 4d is a schematic diagram of a third step diffusion in the investment network according to the embodiment of the present invention.
At this point the destination company is C1, sets an initial score of 1 unit for tags T1 and T2 contained in C1, as shown in FIG. 4a, and then begins the diffusion process. In the first step of diffusion, each label averagely distributes the score of the label to connected investors according to the weight, and the investors add the obtained scores to obtain the score of the label, as shown in fig. 4 b; in the second step of diffusion, each investor distributes all the scores obtained by the investor to the connected companies and labels according to the weight average, wherein lambda times of the total score is distributed to the company set, and 1-lambda times of the total score is distributed to the label set, and at the moment, the companies and the labels obtain corresponding scores, as shown in fig. 4 c; in the third step of diffusion, each company and each label respectively distribute the scores obtained by the company and each label to the connected investors according to the weight average, and the investors add the scores distributed from the two sides to obtain new scores, as shown in fig. 4 d. According to the same rule, the embodiment of the invention can continue to carry out multi-step diffusion, and takes the condition that the score falls in the sponsor set (namely, the score falls in the sponsor set after odd-step diffusion) as a diffusion end point. The optimal number of diffusion steps can be trained from historical data.
In some embodiments of the invention, λ times the value of the sponsor node is for assigning to a sponsor node connected to the sponsor node by a first type connecting edge, λ times the value of the sponsor node (1- λ) is for assigning to a tag node connected to the sponsor node by a second type connecting edge, λ is a value greater than 0 and less than 1, λ is a predetermined optimal bidirectional diffusion specific gravity.
Assuming that each sponsor node has a value of "1", the "1" score is divided and distributed to the left and right. For example, "0.3 parts" is allocated to the left side and "0.7 parts" is allocated to the right side. The reason why the average score is not directly averaged (left "0.5 parts", right "0.5 parts") here is to introduce the parameter λ in consideration that the adjustment of specific gravity spreading to the left and right in the three-part graph has an influence on the recommended effect.
In detail, in the first step of diffusion, each label node distributes its score to the connected investor nodes according to the weight, and the investor nodes add the obtained scores to obtain its score, as shown in fig. 4 b:
In the second step of diffusion, each investor node equally distributes all the scores obtained by itself to the connected company nodes and label nodes according to the weight, wherein λ times of the total score is distributed to the company nodes, and 1- λ times of the total score is distributed to the label nodes, and at this time, the company nodes and the label nodes obtain corresponding scores, as shown in fig. 4 c:
In the third step of diffusion, each company node and each label node respectively distribute the score obtained by each company node to the connected investors according to the weight average, and the investor nodes add the scores distributed from the two sides to obtain a new score, as shown in fig. 4 d:
i1 left
C2 diffusing into
Score, C4 diffusion to
A score value;
i2 left:
c3 diffusing into
Score, C5 spread to
A score value;
i3 left:
c4 diffusing into
Score, C5 spread to
A score value;
i4 left:
c5 diffusing into
A score value;
i1 Right:
t1 diffusing into
Score, T2, is spread to
Score, T3, is spread to
Score, T4, is spread to
And (5) scoring.
I2 Right:
t1 diffusing into
Score, T2, is spread to
Score, T4, is spread to
And (5) scoring.
I3 Right:
t3 diffusing into
Score, T4 diffusion to I3
And (5) scoring.
I4 Right:
t4 diffusion to
And (5) scoring.
40. And determining whether the executed diffusion steps of the diffusion process reach the predetermined optimal diffusion step number or not, and finishing the diffusion process when the executed diffusion steps reach the optimal diffusion step number.
In the embodiment of the present invention, the optimal diffusion step number may be obtained through a pre-training manner, for example, the optimal diffusion step number may be an odd number of values such as 1, 3, or 5, and when the diffusion step number reaches the predetermined optimal diffusion step number in step 30, the above diffusion process is ended, and the following step 50 is triggered to be executed.
50. And obtaining final scores which are respectively possessed by the M investor nodes in the investment network when the diffusion process is finished, sequencing the M investor nodes from large to small according to the final scores which are respectively possessed, and recommending at least one investor node to the target investor node according to a sequencing result.
In the embodiment of the invention, after the diffusion is finished, the value obtained by each investor node reflects the recommendation degree of the investor for the target company, the values are sorted from large to small, and the investor with the top rank is selected to generate the personalized recommendation list for the target company according to the recommendation number requirement. Because the edges of the three-part graph are connected according to the real contact, the investor is connected with the invested company and the invested label. Therefore, the scores are diffused along the connecting edges, and the nodes with indirect connection are all diffused. For example, the initial score is on the label nodes of the target company, in the first diffusion step, the investor nodes investing the labels receive a certain score, and the investor nodes investing more labels receive more scores. Then the investor nodes spread the scores to the company nodes and the label nodes, the company nodes obtaining the scores are invested companies of investors who invest the labels contained in the target company, and the label nodes obtaining the scores are invested labels of investors who invest the labels contained in the target company. That is, the sponsor nodes, the company nodes, and the tag nodes directly related to the indirect and these tags all receive the scores in succession, and then the scores are collected in a sponsor node set, where the odd steps may be a spread of 1, 3, 5, etc. steps. That is, the weak relationship is continuously strengthened through diffusion and finally reflected in the score of the node of the investor, the larger the score is, the stronger the relationship between the investor and the target company is, the more the possibility that the embodiment of the invention judges the investment relationship between the investor and the target company is, that is, the investor with the high score is recommended to the target company.
In some embodiments of the invention, recommending at least one sponsor to the target company according to the ranking results comprises:
and recommending the first S investor nodes in the sequencing result to the target company, wherein S is a positive integer smaller than M.
The value of S may be 1, that is, the sponsor with the highest matching degree with the target company to be recommended is recommended to it, or the value of S may be 2, or 3, or more, so as to recommend multiple sponsors to the target company, which is convenient for subsequent further sponsor selection.
Taking an actual implementation process as an example, the method specifically includes the following steps:
step S01, obtaining historical investment relation data and related label information, including information of investor, invested company, investment amount, share ratio and the like of each historical investment event and labels contained in the company.
S02, training through historical investment data to obtain the optimal step number T * And optimum lambda * 。
Wherein, the historical investment relation data refers to the investment events between a plurality of investors and companies, for example, a certain investor invests a certain company. And the historical investment data comprises historical investment relations and label information of corresponding companies.
At S02, the following detailed diffusion process is performed: the historical investment data is divided into two parts according to the time sequence, wherein the first part is training data, and the second part is testing data. And establishing a company-investor-label three-part graph according to the training data and the related label information, and determining the edge connection weight. For each company that has no investment relationship in the training data, and that has gained investment in the test data, considered as the test company,setting an initial score of 1 for the label contained in the graph, and calculating the result of the score diffused by odd steps (namely steps 1, 3, 5, 7 and the like) according to the continuous edge weight in the three parts of graphs to obtain the recommended scores of all investors. Sequencing investors from large to small according to the recommended values, and calculating sequencing scores according to the ranking of the investors who actually invest the testing company in the testing data
Where r is the rank of the investors who actually invest the company in the test set and L is the total number of investors participating in the ranking.
Changing lambda and diffusion step number continuously, and obtaining the optimal step number T when the average ranking score of all the testing companies is minimum, namely the accuracy is highest * And λ * . The ranking score is a common effect evaluation index of the recommendation system, and besides the ranking score, other accuracy measures such as Recall (Recall), AUC and the like can also be used here. Generally, 3 steps or 5 steps can obtain a result with higher accuracy.
S03, establishing a company-investor-label three-part graph according to all historical investment relation data and related label information, determining the continuous edge weight, establishing an initial score for a label contained in a target company to be recommended, and setting the optimal parameter lambda * Substituting, calculating the diffusion T * And (4) obtaining the recommendation degree of each sponsor for the target user according to the recommendation score of the sponsors after the step, and selecting the sponsors with higher scores to generate a recommendation list of the target company.
The measure of the recommendation effect can also be evaluated by ranking scores, that is, the accuracy of the recommendation can be measured by using investors who actually invest in the target companies, and of course, the embodiment of the invention can also manually pick out some companies, and the aforementioned diffusion process is performed on the companies which invest the companies and the investment events after the companies are unknown, so that the accuracy test is performed after the recommendation result is obtained.
As can be seen from the foregoing illustration, in the embodiment of the present application, an investment network is generated, which includes: m investor nodes, N investor nodes, P label nodes, a first type connecting edge between the investor node and the investor node, and a second type connecting edge between the investor node and the label node. In the embodiment of the invention, the diffusion process can be iteratively executed for a plurality of times, so that the initial values of the label nodes in the diffusion process can be continuously diffused in the investment network, because the connecting edges of the three parts of nodes in the investment network are connected according to the real contact, and the investor is connected with the invested nodes (i.e. invested companies, referred to as companies for short) and the invested labels. Therefore, the scores are diffused along the connecting edges, and the nodes with indirect connection are all diffused. For example, the initial score is on the label nodes of the target company, in the first diffusion step, the investor nodes investing the labels receive a certain score, and the investor nodes investing more labels receive more scores. Then the investor nodes spread the scores to the company nodes and the label nodes, the company nodes obtaining the scores are invested companies of investors who invest the labels contained in the target company, and the label nodes obtaining the scores are invested labels of investors who invest the labels contained in the target company. That is, the sponsor nodes, the corporate nodes, and the tag nodes that directly relate to the indirect and these tags will all receive the scores in succession, and then these scores will be concentrated in the set of sponsor nodes. Therefore, the weak relationship is continuously strengthened through diffusion and finally reflected in the score of the node of the investor, the larger the score is, the stronger the relationship between the investor and the target company is, the higher the possibility of generating investment relationship between the investor and the target company is, namely, the investor with high score is recommended to the target company. Therefore, the accuracy rate of recommending the investor for the initial company can be improved.
The above embodiments are illustrative of the present invention, and are not intended to limit the present invention, and many other embodiments may be equally modified or substituted based on the above embodiments, and are not described in detail herein.
Claims (4)
1. A method for sponsor recommendation, comprising:
obtaining historical investment event information and label information from a historical investment database, wherein the historical investment event information is used for indicating a historical investment relationship between an x-th investor in M investors and an i-th investor in N investors, the label information comprises labels corresponding to the N investors respectively, each investor in the N investors comprises at least one label, the M, the N, the x and the i represent positive integers, the x is smaller than or equal to the M, and the i is smaller than or equal to the N; the historical investment event information includes: the identifier of the xth investor, the identifier of the ith invested party and investment record data; the ith invested party node comprises at least one label used for reflecting the industry field and the main operation business of the ith invested party;
generating an investment network according to the historical investment event information and the label information, wherein the investment network comprises: m investor nodes, N investor nodes, P label nodes, a first type connecting edge between the investor node and the investor node, and a second type connecting edge between the investor node and the label node; the M investor nodes correspond to the M investors one by one, the N investor nodes correspond to the N invested parties one by one, the P label nodes correspond to all different labels corresponding to the N invested parties one by one, the first type connecting edge indicates that an investment relation is established between the x-th investor node and the i-th investor node, and the second type connecting edge indicates that an investment relation is established between the x-th investor node and the label node corresponding to the label included in the i-th investor node;
performing, for a plurality of iterations, a diffusion process comprising: distributing the score obtained by a first type node in the investment network in the current diffusion step to a first neighbor node connected with the first type node, wherein the first neighbor node uses the obtained score in the next diffusion step, label nodes corresponding to all labels included in a target investor node to be recommended in the investment network have initial scores in the diffusion process, and other label nodes except the label nodes corresponding to all labels included in the target investor node in the investment network do not have initial scores in the diffusion process; wherein, when the first type node is the label node, the first neighbor node comprises: the sponsor node is connected with the label node through the second type connecting edge; when the first type node is the sponsor node, the first neighbor node comprises: the label node is connected with the sponsor node through the second type connecting edge, and the sponsor node is connected with the sponsor node through the first type connecting edge; when the first type node is the sponsor node and the tag node, the first neighbor node includes: a sponsor node connected with the sponsor node through the first type connecting edge and a sponsor node connected with the label node through the second type connecting edge;
the diffusion process specifically comprises the following steps:
a, step a: in an initial step before beginning diffusion, setting an initial score for label nodes corresponding to all labels included by target investor nodes to be provided with recommendation in the investment network, and setting scores to be 0 for other label nodes except for label nodes corresponding to all labels included by the target investor nodes in the investment network, wherein the label nodes are M investor nodes, N investor nodes and P label nodes in the investment network;
step b: when the first type node is the label node, the initial score of the label node is averagely distributed to the investor node through the second type connecting edge according to the weight; after the step b is executed, sequentially and iteratively executing the following steps c and d, and stopping iteration until the optimal diffusion step number is reached;
step c: when the first type node is the investor node, the score obtained by the investor node through the step b is averagely distributed to the label node and the investor node through the second type connecting edge and the first type connecting edge according to the weight;
step d: when the first type node is the invested node and the label node, the value obtained by the invested node through the step c and the value obtained by the label node through the step c are averagely distributed to the investor node through the first type connecting edge and the second type connecting edge according to the weight, the investor node superposes the values obtained through the first type connecting edge and the second type connecting edge, and the investor node triggers the execution of the step c through the value obtained by the investor node through the step d;
λ times of the score of the sponsor node is used for being distributed to a sponsor node connected with the sponsor node through the first type connecting edge, 1- λ times of the score of the sponsor node is used for being distributed to a tag node connected with the sponsor node through the second type connecting edge, λ is a numerical value larger than 0 and smaller than 1, and λ is a predetermined optimal bidirectional diffusion specific gravity;
determining whether the executed diffusion steps of the diffusion process have reached a predetermined optimal number of diffusion steps, and ending the diffusion process when the executed diffusion steps have reached the optimal number of diffusion steps;
and obtaining final scores which are respectively possessed by the M investor nodes in the investment network when the diffusion process is finished, sequencing the M investor nodes from large to small according to the final scores which are respectively possessed, and recommending at least one investor node to the target investor node according to a sequencing result.
2. The method of claim 1, wherein after generating an investment network based on said historical investment event information and said tag information, said method further comprises:
determining the weight of the first type connecting edge according to investment record data between the investor node and the invested party node;
and determining the weight of the second type connecting edge according to the investment relation between the investor node and the invested party node, the affiliated relation between the invested party node and the label and the investment record data between the investor node and the invested party node.
3. The method according to claim 2, wherein the assigning the score obtained by the first type node in the investment network in the current flooding step to a first neighboring node connected to the first type node comprises:
and averagely distributing the score obtained by the first type node in the investment network in the current diffusion step to the first neighbor node according to the weight of the first type connecting edge and the weight of the second type connecting edge.
4. The method of claim 1, wherein recommending at least one sponsor node to the target sponsor node according to the ranking result comprises:
recommending the first S investor nodes in the sequencing result to the target investor node, wherein S is a positive integer smaller than M.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910359691.7A CN110264364B (en) | 2019-04-30 | 2019-04-30 | Recommendation method for investor |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910359691.7A CN110264364B (en) | 2019-04-30 | 2019-04-30 | Recommendation method for investor |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110264364A CN110264364A (en) | 2019-09-20 |
| CN110264364B true CN110264364B (en) | 2022-08-23 |
Family
ID=67913949
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910359691.7A Active CN110264364B (en) | 2019-04-30 | 2019-04-30 | Recommendation method for investor |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110264364B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110750653B (en) * | 2019-10-22 | 2023-01-06 | 中国工商银行股份有限公司 | Information processing method, information processing apparatus, electronic device, and medium |
| CN112488842A (en) * | 2020-12-14 | 2021-03-12 | 天津北晟企业服务有限公司 | Investment institution recommendation method and device |
Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103544632A (en) * | 2013-07-22 | 2014-01-29 | 杭州师范大学 | Method and system for individually recommending network commodities |
| US8832091B1 (en) * | 2012-10-08 | 2014-09-09 | Amazon Technologies, Inc. | Graph-based semantic analysis of items |
| CN104899763A (en) * | 2015-05-07 | 2015-09-09 | 西安电子科技大学 | Personalized recommendation method based on bilateral diffusion of bipartite network |
| CN106326367A (en) * | 2016-08-11 | 2017-01-11 | 华南师范大学 | Mixed collaborative recommendation algorithm based on WNBI and RSVD |
| CN107704632A (en) * | 2017-10-31 | 2018-02-16 | 河海大学 | Modification method is recommended based on the Chinese label of synonym and antonym |
| CN108038746A (en) * | 2017-11-24 | 2018-05-15 | 河海大学 | Method is recommended based on the bigraph (bipartite graph) of key user and time context |
| CN108256958A (en) * | 2017-12-22 | 2018-07-06 | 华南师范大学 | A kind of mixing Collaborative Recommendation algorithm based on WUDiff and RMF |
| CN108269117A (en) * | 2017-01-03 | 2018-07-10 | 阿里巴巴集团控股有限公司 | The push of data determines method and device, terminal |
| CN108805642A (en) * | 2017-05-02 | 2018-11-13 | 合信息技术（北京）有限公司 | Recommend method and device |
| CN109165847A (en) * | 2018-08-24 | 2019-01-08 | 广东工业大学 | A kind of item recommendation method based on recommender system, device and equipment |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8589409B2 (en) * | 2010-08-26 | 2013-11-19 | International Business Machines Corporation | Selecting a data element in a network |
| US20180032930A1 (en) * | 2015-10-07 | 2018-02-01 | 0934781 B.C. Ltd | System and method to Generate Queries for a Business Database |
-
2019
- 2019-04-30 CN CN201910359691.7A patent/CN110264364B/en active Active
Patent Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8832091B1 (en) * | 2012-10-08 | 2014-09-09 | Amazon Technologies, Inc. | Graph-based semantic analysis of items |
| CN103544632A (en) * | 2013-07-22 | 2014-01-29 | 杭州师范大学 | Method and system for individually recommending network commodities |
| CN104899763A (en) * | 2015-05-07 | 2015-09-09 | 西安电子科技大学 | Personalized recommendation method based on bilateral diffusion of bipartite network |
| CN106326367A (en) * | 2016-08-11 | 2017-01-11 | 华南师范大学 | Mixed collaborative recommendation algorithm based on WNBI and RSVD |
| CN108269117A (en) * | 2017-01-03 | 2018-07-10 | 阿里巴巴集团控股有限公司 | The push of data determines method and device, terminal |
| CN108805642A (en) * | 2017-05-02 | 2018-11-13 | 合信息技术（北京）有限公司 | Recommend method and device |
| CN107704632A (en) * | 2017-10-31 | 2018-02-16 | 河海大学 | Modification method is recommended based on the Chinese label of synonym and antonym |
| CN108038746A (en) * | 2017-11-24 | 2018-05-15 | 河海大学 | Method is recommended based on the bigraph (bipartite graph) of key user and time context |
| CN108256958A (en) * | 2017-12-22 | 2018-07-06 | 华南师范大学 | A kind of mixing Collaborative Recommendation algorithm based on WUDiff and RMF |
| CN109165847A (en) * | 2018-08-24 | 2019-01-08 | 广东工业大学 | A kind of item recommendation method based on recommender system, device and equipment |
Non-Patent Citations (7)
| Title |
|---|
| Collaborative filtering with diffusion-based similarity on tripartite graphs;M. Shang 等;《Physica A: Statistical Mechanics and its Applications》;20100315;第389卷(第6期);第1259-1264页 * |
| Personalized recommendation via integrated diffusion on user-item-tag tripartite graphs;Zi-Ke Zhang 等;《Physica A: Statistical Mechanics and its Applications》;20100101;第389卷(第1期);第179-186页 * |
| Profiling users with tag networks in diffusion-based personalized recommendation;Jin Mao 等;《Journal of Information Science》;20151019;第42卷(第5期);第711-722页 * |
| Solving the cold-start problem in recommender systems with social tags;Z.-K. Zhang 等;《Europhysics Letters》;20100607;第92卷(第2期);第28002（1-7）页 * |
| 一种基于用户信任网络的物质扩散推荐算法;梁艺 等;《计算机与现代化》;20180215(第02期);第12-16页 * |
| 基于加权三部图模型的推荐算法研究;戴瑾如;《中国优秀硕士学位论文全文数据库 信息科技辑》;20190115;第2019年卷(第01期);第I138-5450页 * |
| 结合三部图综合扩散的Slope One推荐算法;王冉 等;《微电子学与计算机》;20180405;第35卷(第04期);第6-11页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110264364A (en) | 2019-09-20 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Dasgupta et al. | Inattentive importers | |
| Buskens | Social networks and trust | |
| US6873979B2 (en) | Method of building predictive models on transactional data | |
| Li et al. | Price to compete… with many: How to identify price competition in high-dimensional space | |
| Xie et al. | Incentive mechanism and rating system design for crowdsourcing systems: Analysis, tradeoffs and inference | |
| CN111798280B (en) | Multimedia information recommendation method, device and equipment and storage medium | |
| Orr | Within-firm productivity dispersion: Estimates and implications | |
| CN110264364B (en) | Recommendation method for investor | |
| CN112445699A (en) | Strategy matching method and device, electronic equipment and storage medium | |
| CN106202151A (en) | One is used for improving the multifarious method of personalized recommendation system | |
| Xie et al. | Incentive and reputation mechanisms for online crowdsourcing systems | |
| Park | Factors affecting consumers’ intention to use online music service and customer satisfaction in South Korea | |
| Xu et al. | Predicting advertiser bidding behaviors in sponsored search by rationality modeling | |
| Shang et al. | Need for speed, but how much does it cost? Unpacking the fee‐speed relationship in Bitcoin transactions | |
| Xuefeng et al. | Predicting the final prices of online auction items | |
| CN108596765A (en) | A kind of Electronic Finance resource recommendation method and device | |
| CN109075987A (en) | Optimize digital assembly analysis system | |
| Hyde et al. | Investigating concurrency in online auctions through visualization | |
| CN112581281A (en) | Product recommendation method and device, storage medium and electronic equipment | |
| Arsad et al. | Selection input output by restriction using DEA models based on a fuzzy Delphi approach and expert information | |
| Shi | The application of machine learning in online purchasing intention prediction | |
| Phulara et al. | A novel hybrid grey-BCM approach in multi-criteria decision making: An application in OTT platform | |
| CN116361542A (en) | Product recommendation method, device, computer equipment and storage medium | |
| Purba et al. | Decision Making for Optimal Marketing Strategy: A Case Study in E-Commerce | |
| Li et al. | Information acquisition in new product introduction |
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 |