Data processing method, device and electronic equipment
By optimizing the vector elements of the asset object, the problem of high noise in asset similarity calculation is solved, and the accuracy is improved, especially in the case of missing values in the daily yield sequence.
Patent Information
- Application Number
- CN202210061843.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-01-19
- Publication Date
- 2024-11-05
- Estimated Expiration
- 2042-01-19
AI Technical Summary
In the prior art, there is high noise in the calculation of asset similarity, resulting in low accuracy, especially in the case of missing values in the daily rate of return sequence of assets.
By obtaining the object similarity matrix of the asset object collection, the vector elements of the asset object are optimized using the similarity error function until the iterative convergence conditions are met, and the optimized vector elements are obtained to improve the accuracy of the asset similarity.
It effectively improves the accuracy of asset similarity, avoids the problem of low accuracy due to the lack of yield, and achieves higher similarity calculation accuracy.
Smart Images

Figure CN114418762B_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of data technology, and in particular to a data processing method, device and electronic device. Background Art
[0002] In the financial field, any stock, fund, index, etc. can be called an asset, and each asset has a corresponding daily yield sequence data. In order to analyze the relationship between assets, it is necessary to quantify the asset data, calculate and analyze the similarity between each asset, and cluster the assets, so as to facilitate subsequent analysis and strategy research on similar assets.
[0003] At present, assets are usually represented by their daily yield series data, and the similarity between assets is obtained by calculating the correlation coefficient between the assets in the daily yield series.
[0004] However, since the daily yield sequence of an asset is inherently random, there may be cases where daily yields are missing, resulting in the obtained correlation coefficient having greater noise, making the accuracy of obtaining asset similarity lower. Summary of the invention
[0005] In view of this, the present application provides a data processing method, device and electronic device to solve the technical problem that the accuracy of asset similarity obtained by the prior art is low. As follows:
[0006] A data processing method, the method comprising:
[0007] Obtain an asset object set, wherein the asset object set includes a plurality of asset objects, the asset objects have object data, the asset objects are represented by object vectors, the object vectors include a plurality of vector elements, and the vector elements have initial element values;
[0008] According to the object data of the asset objects, an object similarity matrix corresponding to the asset object set is obtained, wherein the matrix elements of the object similarity matrix are the initial similarities between any two asset objects in the asset object set;
[0009] According to the matrix elements in the object similarity matrix, the initial element values of the vector elements in the object vector of the asset object are processed to obtain the optimized vector elements in the object vector of the asset object, and the vector elements in the object vector are used to obtain the asset similarity between the asset object to which it belongs and other asset objects.
[0010] The above method preferably processes the initial element values of the vector elements in the object vector of the asset object according to the matrix elements in the object similarity matrix to obtain the optimized vector elements in the object vector of the asset object, including:
[0011] Taking the initial similarity in the matrix elements in the object similarity matrix as the input of the similarity error function, and optimizing the initial element values of the vector elements in the object vector of the asset object by using the similarity error function, so as to obtain the optimized vector elements in the object vector of the asset object;
[0012] The similarity error function is represented by at least the sum of the similarity errors between all any two asset objects in the asset object set, and the similarity error between the asset objects in the similarity error function is represented by at least the vector distance between the object vectors of the corresponding asset objects and the initial similarity between the corresponding asset objects.
[0013] The above method preferably uses the initial similarity of the matrix elements in the object similarity matrix as the input of the similarity error function, and uses the similarity error function to optimize the initial element values of the vector elements in the object vector of the asset object, including:
[0014] Using the initial similarity in the matrix elements in the object similarity matrix as the input of the similarity error function, obtaining a first output function value of the similarity error function for the initial element value of the vector element in the object vector of the asset object;
[0015] According to the first output function value, adjusting the initial element value of the vector element in the object vector of the asset object;
[0016] Taking the initial similarity in the matrix elements in the object similarity matrix as the input of the similarity error function, obtaining a second output function value of the similarity error function for the adjusted element value of the vector element in the object vector of the asset object;
[0017] At least according to the second output function value, adjust the adjusted element value of the vector element in the object vector of the asset object, and return to execute the step of: using the initial similarity in the matrix element in the object similarity matrix as the input of the similarity error function, obtaining the second output function value of the similarity error function for the adjusted element value of the vector element in the object vector of the asset object, until the adjusted element value of the vector element in the object vector of the asset object satisfies the iterative convergence condition, so as to obtain the optimized vector element in the object vector of the asset object;
[0018] The iteration convergence condition is that the change in the adjusted element value of the vector element in the object vector of the asset object is less than or equal to a change threshold.
[0019] In the above method, preferably, in the similarity error between the two asset objects in the similarity error function, the initial similarity between the two asset objects is subjected to logarithm processing;
[0020] The similarity error between the two asset objects in the similarity error function is also represented by vector offset items corresponding to the object vectors of the two asset objects, and the vector offset items are related to the vector elements in the object vectors corresponding thereto;
[0021] The similarity error between the two asset objects in the similarity error function is squared;
[0022] The similarity error between the two asset objects in the similarity error function is weighted.
[0023] In the above method, preferably, the object data includes yield data corresponding to multiple dates;
[0024] The matrix elements of the object similarity matrix are obtained in the following manner:
[0025] Obtain the yield difference of the object data of the two asset objects corresponding to the matrix elements on each date;
[0026] At least according to the yield difference on each of the dates, the initial similarity between the two asset objects corresponding to the matrix elements is obtained.
[0027] The above method preferably obtains the initial similarity between the two asset objects corresponding to the matrix elements at least according to the yield difference on each of the dates, including:
[0028] The yield difference of the object data of the two asset objects corresponding to the matrix elements on each date is processed respectively to obtain the similarity component corresponding to each date;
[0029] The similarity components are accumulated according to the dates to obtain an initial similarity between the two asset objects corresponding to the matrix elements.
[0030] The above method preferably processes the yield difference of the object data of the two asset objects corresponding to the matrix elements on each date respectively to obtain the similarity component corresponding to each date, including:
[0031] Respectively obtain the absolute value of the yield difference between the object data of the two asset objects corresponding to the matrix elements on each date;
[0032] The absolute values of the yield differences are processed respectively to obtain the similarity components corresponding to each of the dates, and the similarity components corresponding to each of the dates are values greater than or equal to 0.
[0033] The above method, preferably, further comprises:
[0034] Obtaining vector elements in object vectors of any plurality of the asset objects in the asset object set;
[0035] According to the vector elements in the object vectors of the arbitrary multiple asset objects, the cosine similarity between the object vectors of the arbitrary multiple asset objects is obtained; and the cosine similarity is used as the asset similarity between the arbitrary multiple asset objects.
[0036] A data processing device, comprising:
[0037] An object obtaining unit, used for obtaining an asset object set, wherein the asset object set includes a plurality of asset objects, the asset objects have object data, the asset objects are represented by object vectors, the object vectors include a plurality of vector elements, and the vector elements have initial element values;
[0038] A matrix obtaining unit, configured to obtain an object similarity matrix corresponding to the asset object set according to the object data of the asset object, wherein the matrix elements of the object similarity matrix are initial similarities between any two asset objects in the asset object set;
[0039] A vector optimization unit is used to process the initial element values of the vector elements in the object vector of the asset object according to the matrix elements in the object similarity matrix to obtain the optimized vector elements in the object vector of the asset object, and the vector elements in the object vector are used to obtain the asset similarity between the asset object to which it belongs and other asset objects.
[0040] An electronic device, comprising:
[0041] A memory, used to store applications and data generated by the operation of the applications;
[0042] A processor is used to execute the application to achieve: obtaining an asset object set, wherein the asset object set includes multiple asset objects, the asset objects have object data, the asset objects are represented by object vectors, the object vectors include multiple vector elements, and the vector elements have initial element values; obtaining an object similarity matrix corresponding to the asset object set according to the object data of the asset objects, wherein the matrix elements of the object similarity matrix are the initial similarities between any two asset objects in the asset object set; and processing the initial element values of the vector elements in the object vector of the asset object according to the matrix elements in the object similarity matrix to obtain optimized vector elements in the object vector of the asset object, wherein the vector elements in the object vector are used to obtain the asset similarity between the asset object to which it belongs and other asset objects.
[0043] It can be seen from the above technical scheme that in a data processing method, device and electronic device disclosed in the present application, the object data of the asset object is used to obtain the object similarity matrix corresponding to the asset object set to which it belongs, and the matrix elements in the object similarity matrix are the initial similarities between any two asset objects in the asset object set. Based on this, according to the initial similarity in the object similarity matrix, the vector elements in the object vector representing the asset object are optimized, thereby, the asset object is represented by the object vector composed of the optimized vector elements, and then the asset similarity between the asset objects can be obtained based on the object vector of the asset object. It can be seen that in the present application, after the initial similarity between all any two asset objects is obtained by using the object data of the asset object, the vector elements of all asset objects represented by vectorization are optimized by using the initial similarity, thereby, the asset object is represented by the optimized object vector, thereby avoiding the situation where the accuracy of the asset similarity obtained by relying solely on the data such as the yield of two asset objects is low, and the similarity between the asset objects is obtained by the optimized object vector, thereby achieving the purpose of improving the accuracy of the asset similarity. BRIEF DESCRIPTION OF THE DRAWINGS
[0044] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.
[0045] Figure 1 A flowchart of a data processing method provided in Example 1 of the present application;
[0046] Figure 2-Figure 4 They are respectively partial flow charts of a data processing method provided in Example 1 of the present application;
[0047] Figure 5 Another flow chart of a data processing method provided in Embodiment 1 of the present application;
[0048] Figure 6 A schematic diagram of the structure of a data processing device provided in Embodiment 2 of the present application;
[0049] Figure 7 Another structural schematic diagram of a data processing device provided in Embodiment 2 of the present application;
[0050] Figure 8 A schematic diagram of the structure of an electronic device provided in Embodiment 3 of the present application;
[0051] Fig. 9 This is a schematic diagram of an asset vector applicable to asset similarity acquisition according to an embodiment of the present application. DETAILED DESCRIPTION
[0052] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.
[0053] refer to Figure 1 , is a flowchart of a data processing method provided in the first embodiment of the present application, which can be applied to electronic devices capable of data processing, such as computers or servers. The technical solution in this embodiment is mainly used to improve the accuracy of asset similarity obtained based on vectors through vectorized representation of asset objects.
[0054] Specifically, the method in this embodiment may include the following steps:
[0055] Step 101: Obtain a collection of asset objects.
[0056] The asset object set includes multiple asset objects, and the asset objects have object data. The object data here may include yield data corresponding to multiple dates, which may be specifically represented by a daily yield sequence.
[0057] In this embodiment, each asset object in the asset object set is represented by an object vector. The object vector of the asset object may contain multiple vector elements, each vector element corresponds to a different vector dimension, and the element value of the vector element in each vector dimension may be the same or different. In addition, in this embodiment, the vector elements in the object vector of the asset object are set with initial element values. For example, in this embodiment, the object vector of the asset object is set to 100 vector dimensions, and the initial element values of the vector elements in these 100 vector dimensions are calculated by a random algorithm.
[0058] Step 102: Obtain an object similarity matrix corresponding to the asset object set according to the object data of the asset object, wherein the matrix elements of the object similarity matrix are the initial similarities between any two asset objects in the asset object set.
[0059] In this embodiment, the object data of any two asset objects in the asset object set can be processed to obtain the initial similarity between any two asset objects in the asset object set, and then these initial similarities are set as matrix elements in the object similarity matrix. Based on this, the rows and columns of the object similarity matrix are respectively the total amount of asset objects in the asset object set, and the object similarity matrix is a symmetric matrix.
[0060] Taking an asset object set including N asset objects as an example, in this embodiment, an object similarity matrix is constructed for these N asset objects, which can also be called an asset similarity matrix and can be represented by S. The object similarity matrix is a matrix of size N*N, where N is the total number of asset objects. The object similarity matrix is a symmetric matrix, that is, the matrix element S[i, j] in the i-th row and j-th column is the same as the matrix element S[j, i] in the j-th row and i-th column, where i is not equal to j. Specifically, each matrix element in the object similarity matrix is the initial similarity between the two asset objects obtained based on the object data of the two asset objects corresponding to the matrix element.
[0061] Step 103: processing the initial element values of the vector elements in the object vector of the asset object according to the matrix elements in the object similarity matrix to obtain the optimized vector elements in the object vector of the asset object.
[0062] The vector elements in the object vector are used to obtain the asset similarity between the asset object to which it belongs and other asset objects.
[0063] It should be noted that since the object data of a certain asset object may be missing, the initial similarity between asset objects obtained based on the object data may be inaccurate. Therefore, in this embodiment, the asset object is represented by an object vector, and the initial similarity between asset objects is used to optimize the initial element values of the vector elements in the object vector. Thus, the asset object is represented by an object vector with optimized vector elements. Based on this, the asset similarity between asset objects obtained based on the optimized vector elements has higher accuracy than the initial similarity.
[0064] Specifically, in this embodiment, after constructing a vector similarity with the object vectors of any two asset objects in the asset object set, a similarity error function is constructed based on the vector similarity and the corresponding initial similarity, and then the error value between the vector similarity and the initial pixel degree of the similarity between any two asset objects output by the similarity error function is minimized, thereby obtaining an object vector composed of vector elements with optimized element values, so as to optimize the object vector to represent the corresponding asset object.
[0065] It can be seen from the above scheme that in a data processing method provided in the first embodiment of the present application, the object data of the asset object is used to obtain the object similarity matrix corresponding to the asset object set to which it belongs, and the matrix elements in the object similarity matrix are the initial similarities between any two asset objects in the asset object set. Based on this, according to the initial similarity in the object similarity matrix, the vector elements in the object vector representing the asset object are optimized, thereby representing the asset object with the object vector composed of the optimized vector elements, and then the asset similarity between the asset objects can be obtained based on the object vector of the asset object. It can be seen that in this embodiment, after the initial similarity between all any two asset objects is obtained using the object data of the asset object, the vector elements of all asset objects represented by vectorization are optimized using the initial similarity, thereby, the asset object is represented by the optimized object vector, thereby avoiding the situation where the accuracy of the asset similarity obtained by relying solely on the data such as the yield of two asset objects is low, and the similarity between the asset objects is obtained through the optimized object vector, thereby achieving the purpose of improving the accuracy of the asset similarity.
[0066] In one implementation, in step 103, when the initial element values of the vector elements in the object vector of the asset object are processed according to the matrix elements in the object similarity matrix, it can be specifically implemented in the following manner:
[0067] Taking the initial similarity in the matrix elements in the object similarity matrix as the input of the similarity error function, and optimizing the initial element values of the vector elements in the object vector of the asset object by using the similarity error function, so as to obtain the optimized vector elements in the object vector of the asset object;
[0068] The similarity error function is represented by at least the sum of the similarity errors between all any two asset objects in the asset object set, wherein the similarity error between the asset objects in the similarity error function is represented by at least the vector distance between the object vectors of the corresponding asset objects and the initial similarity between the corresponding asset objects.
[0069] Specifically, the similarity error function includes multiple error terms, the sum of these error terms constitutes the similarity error function, and each error term is the similarity error between all any two asset objects in the asset object set. Specifically, each error term is determined by the vector distance between the object vectors of the corresponding asset objects and the initial similarity between the corresponding asset objects, wherein the vector distance between the object vectors of the asset objects can be understood as the vector similarity between the object vectors of the asset objects, wherein the initial similarity between the asset objects can be obtained from the corresponding matrix elements in the object similarity matrix.
[0070] Therefore, the output function value of the similarity error function represents the sum of the errors between the vector similarity and the initial similarity of all two arbitrary asset objects in the asset object set.
[0071] Since the vector distance in the error term included in the similarity error function is represented by the initial element value of the vector element in the object vector of the corresponding asset object, and the initial similarity in the error term included in the similarity error function is determined, based on this, in this embodiment, the initial similarity in the object similarity matrix can be used to minimize the output of the similarity error function, and then through multiple iterations of optimization processing, the element values of the vector elements in the object vector of the asset object are gradually adjusted, so as to obtain the element value of the optimized vector element when the similarity error between all two arbitrary asset objects is the smallest. Therefore, in this embodiment, the similarity between asset objects can be obtained through the optimized object vector, thereby achieving the purpose of improving the accuracy of asset similarity.
[0072] The following describes the process of optimizing the initial element values of the vector elements in the object vector of the asset object using the similarity error function. Figure 2 As shown in:
[0073] Step 201: using the initial similarity in the matrix elements in the object similarity matrix as the input of the similarity error function, obtaining the first output function value of the similarity error function for the initial element value of the vector element in the object vector of the asset object.
[0074] Among them, in this embodiment, the initial similarities of all matrix elements in the object similarity matrix are input into the error term of the similarity error function, and the initial element values of the vector elements in the object vectors of all asset objects are also input into the error term of the similarity error function. Thus, the similarity error function is calculated to obtain the first output function value of the similarity error function.
[0075] Step 202: According to the first output function value, the initial element value of the vector element in the object vector of the asset object is adjusted.
[0076] In this embodiment, the initial element value of the vector element in the object vector of the asset object can be increased or decreased according to the size of the first output function value, thereby obtaining the adjusted element value of the vector element in the object vector of the asset object.
[0077] Step 203: using the initial similarity in the matrix elements in the object similarity matrix as the input of the similarity error function, obtaining a second output function value of the similarity error function for the adjusted element value of the vector element in the object vector of the asset object.
[0078] Among them, in this embodiment, the initial similarities of all matrix elements in the object similarity matrix can be input into the error term of the similarity error function, and the adjusted element values of the vector elements in the object vectors of all asset objects can be input into the error term of the similarity error function. Thus, the similarity error function is calculated again to obtain the second output function value of the similarity error function.
[0079] Step 204: Adjust the adjusted element values of the vector elements in the object vector of the asset object at least according to the second output function value, return to execute step 203, and again use the initial similarity in the matrix elements in the object similarity matrix as the input of the similarity error function to obtain the second output function value of the similarity error function for the adjusted element values of the vector elements in the object vector of the asset object, until the adjusted element values of the vector elements in the object vector of the asset object meet the iterative convergence condition to obtain the optimized vector elements in the object vector of the asset object.
[0080] In this embodiment, the initial element value of the vector element in the object vector of the asset object can be increased or decreased according to the size of the second output function value, thereby obtaining the adjusted element value of the vector element in the object vector of the asset object.
[0081] In an optimized processing method, in this embodiment, based on the size of the second output function value and with reference to the previous adjustment method of the vector elements in the object vector of the asset object, the adjusted element values of the vector elements in the object vector of the asset object can be increased or decreased again, thereby obtaining the re-adjusted element values of the vector elements in the object vector of the asset object.
[0082] It should be noted that the iteration convergence condition is that the change in the element value of the vector element in the object vector of the asset object after adjustment is less than or equal to the change threshold.
[0083] The change in the element value here can be understood as: the difference obtained by subtracting the adjusted element value of the vector element in the object vector of the asset object from the element value adjusted last time, and the difference is used to represent the change in the adjusted element value of the vector element in the object vector of the asset object.
[0084] That is to say, in this embodiment, the initial similarity in the matrix elements in the object similarity matrix is used, and the vector elements in the object vector of the asset object in the error term contained in the similarity error function are iteratively optimized multiple times, so that the total error between all any two asset objects in the asset object set in the vector similarity and the initial similarity is minimized, thereby obtaining the optimal object vector of the asset object.
[0085] For example, with V i An object vector representing the i-th asset object, expressed as V j An object vector representing the j-th asset object, V i The transpose of V j Represents the vector similarity between the i-th asset object and the j-th asset object, with the matrix element S in the object similarity matrix ij Represents the initial similarity between the i-th asset object and the j-th asset object. At this time, the error term between the i-th asset object and the j-th asset object in the similarity error function is V i The transpose of V j Then subtract S ij Indicates that, therefore, the matrix element S in the object similarity matrix ij is the input of the similarity error function, with V i and V j The initial element value of the vector element in is initial. Through multiple iterations, Vi and V j The element values of the vector elements in V are optimized until i and V j The amount by which the element values of the vector elements in change is below a threshold.
[0086] Furthermore, in order to speed up iterative optimization and improve the accuracy of iterative optimization, in this embodiment, any one or more of the following adjustments are made to the similarity error function:
[0087] In one implementation, in the similarity error between two asset objects in the similarity error function, the initial similarity between the two asset objects is processed by taking the logarithm. For example, the error term between the i-th asset object and the j-th asset object in the similarity error function is V i The transpose of V j Then subtract S ij The numerical representation obtained by taking the logarithm;
[0088] In one implementation, the similarity error between two asset objects in the similarity error function is also represented by vector offsets corresponding to the object vectors of the two asset objects, and the vector offsets are related to the vector elements in the corresponding object vectors. i The vector bias term is b i Indicates that V j The vector bias term is b j The error term between the i-th asset object and the j-th asset object in the similarity error function is expressed as follows: V i The transpose of V j After that, add b i and b j Then subtract S ij The value obtained by taking the logarithm;
[0089] In one implementation, the similarity error between two asset objects in the similarity error function is squared. For example, the square of the error term between the i-th asset object and the j-th asset object in the similarity error function is used as the error term;
[0090] In one implementation, the similarity error between two asset objects in the similarity error function is weighted. For example, the error term between the ith asset object and the jth asset object in the similarity error function is multiplied by the corresponding weight value between the ith asset object and the jth asset object, and the corresponding weight value between the ith asset object and the jth asset object is calculated by the initial similarity S between the ith asset object and the jth asset object. ij Sure.
[0091] For example, the corresponding weight value between the i-th asset object and the j-th asset object is obtained in the following way:
[0092] The initial similarity S between the i-th asset object and the j-th asset object ij If the maximum similarity value is less than the preset maximum similarity value, firstly take the exponent of the maximum similarity value to the preset value α, and then add S ij Divide by the value obtained by taking the exponent of the maximum value of the sum of similarities to the preset value α, thereby obtaining the corresponding weight value between the i-th asset object and the j-th asset object;
[0093] The initial similarity S between the i-th asset object and the j-th asset object ij When the similarity sum is less than the preset maximum value, 1 is used as the corresponding weight value between the i-th asset object and the j-th asset object.
[0094] The maximum value of the sum of similarities is determined based on the number of dates involved in the object data of the asset object. For example, if the object data contains yield data for 20 dates, the maximum value of the sum of similarities can be set to 20 or a value thereabout, and so on.
[0095] Specifically, the similarity error function is expressed by the following formula (1):
[0096]
[0097] Among them, f(S ij ) is the corresponding weight value between the i-th asset object and the j-th asset object, expressed as the following formula (2):
[0098]
[0099] Among them, S max is the maximum value of the sum of similarities, where S max It can also be other values, which are related to the number of dates, and α is the preset value.
[0100] In one implementation, the object data of the asset object may include yield data corresponding to multiple dates, wherein the yield data is usually a value less than 0.1.
[0101] Based on this, the matrix elements of the object similarity matrix in step 102 can be obtained by the following method: Figure 3 As shown in:
[0102] Step 301: Obtain the yield difference of the object data of two asset objects corresponding to the matrix elements on each date.
[0103] In this embodiment, the yield data on each date in the object data of the two asset objects corresponding to the matrix elements may be subtracted to obtain the yield difference on each date.
[0104] For example, with r i,t Represents the yield data of the i-th asset object on the t-th date, expressed as r j,t represents the yield data of the jth asset object on the tth date, and the yield difference on the tth date is expressed as r i,t Subtract r j,t express.
[0105] Step 302: Obtain the initial similarity between two asset objects corresponding to the matrix elements at least according to the yield difference on each date.
[0106] In this embodiment, the initial similarity between the two asset objects corresponding to the matrix elements can be obtained by numerically calculating the yield difference. The specific implementation method is as follows Figure 4 As shown in:
[0107] Step 401: Process the yield difference of the object data of the two asset objects corresponding to the matrix elements on each date respectively to obtain the similarity component corresponding to each date.
[0108] In this embodiment, the yield difference on each date is numerically calculated to obtain the similarity component corresponding to each date, and the similarity component needs to be a value greater than or equal to 0. Based on this, in step 401, the absolute value of the yield difference between the object data of the two asset objects corresponding to the matrix elements on each date can be obtained respectively. For example, for r i,t Subtract r j,t The obtained difference is taken as the absolute value; then, the absolute value of the yield difference corresponding to each date is processed separately so that the corresponding similarity component on each date is a value greater than or equal to 0.
[0109] Specifically, first, the absolute value of the yield difference on each date is multiplied by the adjustment coefficient to increase the order of magnitude of the absolute value. Then, 1 is subtracted from the absolute value multiplied by the adjustment coefficient. Finally, it is determined whether the absolute value of user 1 subtracted from the adjustment coefficient is greater than 0. If it is greater than 0 or equal to 0, the value obtained by subtracting the absolute value of user 1 subtracted from the adjustment coefficient is used as the similarity component corresponding to the corresponding date. If it is less than 0, 0 is used as the similarity component corresponding to the corresponding date.
[0110] For example, since the yield data is a value less than 0.1, after obtaining the absolute value of the difference, in order to improve the accuracy, first r i,tSubtract r j,t The absolute value of the difference obtained is multiplied by an adjustment coefficient, such as 10, to increase the order of magnitude of the absolute value of the difference. Then, the absolute value after multiplication by 10 is subtracted from 1, and the maximum value is selected from the difference between 0 and 1 minus the absolute value after multiplication by 10 as the similarity component on the t-th date.
[0111] Step 402: Accumulate the similarity components according to the date to obtain the initial similarity between the two asset objects corresponding to the matrix elements.
[0112] Specifically, in this embodiment, the similarity components on all dates are accumulated to obtain the matrix element, that is, the initial similarity between the corresponding two asset objects.
[0113] For example, the similarity components corresponding to the ith asset object and the jth asset reference on dates 1-20 are accumulated to obtain the initial similarity between the ith asset object and the jth asset reference.
[0114] Specifically, the initial similarity in the matrix elements is S ij It is expressed as the following formula (3):
[0115]
[0116] Here, t starts from 1, and T is the total number of dates. The adjustment factor 10 here can also take other values.
[0117] It can be seen that in the embodiment of the present application, after the initial similarity between any two asset objects is obtained by using the yield data of the asset objects on multiple dates, the vector elements of all asset objects represented by vectorization are optimized using the initial similarity, thereby representing the asset objects with the optimized object vectors. Different from the situation in the prior art where the asset similarity between asset objects is obtained by relying solely on the yields of two asset objects, the embodiment of the present application does not ignore the influence of the yields of other asset objects on the obtained asset similarity, but instead uses the yields of all asset objects in the optimization calculation of the object vectors of the asset objects, and then obtains the similarity between the asset objects through the optimized object vectors, thereby avoiding ignoring the influence of the yields of other asset objects on the asset similarity calculation or the defect of low asset similarity accuracy caused by missing yields, thereby achieving the purpose of improving the accuracy of asset similarity.
[0118] In one implementation, after step 103, the method in this embodiment may further include the following steps: Figure 5 As shown in:
[0119] Step 104: Obtain vector elements in the object vectors of any plurality of asset objects in the asset object set.
[0120] Specifically, in this embodiment, optimized element values of vector elements in object vectors of any plurality of asset objects are obtained.
[0121] Step 105: Obtain the cosine similarity between the object vectors of any multiple asset objects according to the vector elements in the object vectors of any multiple asset objects; and use the cosine similarity as the asset similarity between any multiple asset objects.
[0122] In this embodiment, the cosine similarity calculation can be performed on the object vectors of any multiple asset objects according to the vector elements contained therein, and the obtained cosine similarity can be used to represent the asset similarity between any multiple asset objects.
[0123] It should be noted that, in this embodiment, the cosine similarity of the object vectors of any two asset objects can be calculated to obtain the asset similarity between the two asset objects; or, in this embodiment, the cosine similarity of the object vectors of all asset objects in the asset object set can be calculated to obtain the asset similarity between all asset objects in the asset object set. It can be seen that in this application, the asset objects are represented by vectorization, and then the cosine similarity between vectors is calculated to achieve the global similarity between any multiple asset objects. Furthermore, in this embodiment, the object vectors of the asset objects are optimized based on the initial similarity obtained from the yield data to minimize the similarity error between all asset objects, thereby optimizing the similarity between assets globally, thereby improving the accuracy of the similarity.
[0124] refer to Figure 6 , is a structural diagram of a data processing device provided in Embodiment 2 of the present application, the device can be configured in an electronic device capable of data processing, such as a computer or a server. The technical solution in this embodiment is mainly used to improve the accuracy of asset similarity obtained based on vectors through vectorized representation of asset objects.
[0125] Specifically, the device in this embodiment may include the following units:
[0126] The object obtaining unit 601 is used to obtain an asset object set, wherein the asset object set includes multiple asset objects, each of which has object data, and each of which is represented by an object vector, wherein the object vector includes multiple vector elements, and each of which has an initial element value;
[0127] A matrix obtaining unit 602 is used to obtain an object similarity matrix corresponding to the asset object set according to the object data of the asset object, wherein the matrix elements of the object similarity matrix are the initial similarities between any two asset objects in the asset object set;
[0128] The vector optimization unit 603 is used to process the initial element values of the vector elements in the object vector of the asset object according to the matrix elements in the object similarity matrix to obtain the optimized vector elements in the object vector of the asset object. The vector elements in the object vector are used to obtain the asset similarity between the asset object to which it belongs and other asset objects.
[0129] It can be seen from the above scheme that in a data processing device of the second embodiment of the present application, the object data of the asset object is used to obtain the object similarity matrix corresponding to the asset object set to which it belongs, and the matrix elements in the object similarity matrix are the initial similarities between any two asset objects in the asset object set. Based on this, according to the initial similarity in the object similarity matrix, the vector elements in the object vector representing the asset object are optimized, thereby representing the asset object with the object vector composed of the optimized vector elements, and then the asset similarity between the asset objects can be obtained based on the object vector of the asset object. It can be seen that in this embodiment, after the initial similarity between all any two asset objects is obtained using the object data of the asset object, the vector elements of all asset objects represented by vectorization are optimized using the initial similarity, thereby, the asset object is represented by the optimized object vector, thereby avoiding the situation where the accuracy of the asset similarity obtained by relying solely on the data such as the yield of two asset objects is low, and the similarity between the asset objects is obtained through the optimized object vector, thereby achieving the purpose of improving the accuracy of the asset similarity.
[0130] In one implementation, the vector optimization unit 603 is specifically used to: use the initial similarity in the matrix elements in the object similarity matrix as the input of the similarity error function, and use the similarity error function to optimize the initial element values of the vector elements in the object vector of the asset object to obtain the optimized vector elements in the object vector of the asset object;
[0131] The similarity error function is represented by at least the sum of the similarity errors between all any two asset objects in the asset object set, wherein the similarity error between the asset objects in the similarity error function is represented by at least the vector distance between the object vectors of the corresponding asset objects and the initial similarity between the corresponding asset objects.
[0132] Further, the vector optimization unit 603 is specifically used to: use the initial similarity in the matrix elements in the object similarity matrix as the input of the similarity error function to obtain a first output function value of the similarity error function for the initial element value of the vector element in the object vector of the asset object; adjust the initial element value of the vector element in the object vector of the asset object according to the first output function value; use the initial similarity in the matrix elements in the object similarity matrix as the input of the similarity error function to obtain a second output function value of the similarity error function for the adjusted element value of the vector element in the object vector of the asset object; adjust the adjusted element value of the vector element in the object vector of the asset object at least according to the second output function value, and return to the execution step: use the initial similarity in the matrix elements in the object similarity matrix as the input of the similarity error function to obtain the second output function value of the similarity error function for the adjusted element value of the vector element in the object vector of the asset object, until the adjusted element value of the vector element in the object vector of the asset object meets the iterative convergence condition, so as to obtain the optimized vector element in the object vector of the asset object;
[0133] The iteration convergence condition is that the change in the adjusted element value of the vector element in the object vector of the asset object is less than or equal to the change threshold.
[0134] Preferably, in the similarity error between the two asset objects in the similarity error function, the initial similarity between the two asset objects is processed by taking logarithms;
[0135] The similarity error between two asset objects in the similarity error function is also represented by vector bias items corresponding to the object vectors of the two asset objects, and the vector bias items are related to the vector elements in the object vectors corresponding thereto;
[0136] The similarity error between two asset objects in the similarity error function is squared;
[0137] The similarity error between two asset objects in the similarity error function is weighted.
[0138] In one implementation, the object data includes yield data corresponding to multiple dates;
[0139] Based on this, the matrix obtaining unit 602 obtains the matrix elements of the object similarity matrix in the following manner: obtaining the yield difference between the object data of the two asset objects corresponding to the matrix elements on each date; and obtaining the initial similarity between the two asset objects corresponding to the matrix elements at least based on the yield difference on each date.
[0140] Based on this, when the matrix obtaining unit 602 obtains the initial similarity between the two asset objects corresponding to the matrix elements based on at least the yield difference on each date, it is specifically used to: process the yield difference of the object data of the two asset objects corresponding to the matrix elements on each date respectively to obtain the similarity component corresponding to each date, for example, obtain the absolute value of the yield difference of the object data of the two asset objects corresponding to the matrix elements on each date respectively, and then process the absolute value of the yield difference respectively to obtain the similarity component corresponding to each date, and the similarity component corresponding to each date is a value greater than or equal to 0; thereafter, accumulate the similarity components according to the date to obtain the initial similarity between the two asset objects corresponding to the matrix elements.
[0141] In one implementation, the device in this embodiment may further include the following units: Figure 7 As shown in:
[0142] The similarity acquisition unit 604 is used to obtain vector elements in the object vectors of any multiple asset objects in the asset object set; obtain cosine similarities between the object vectors of any multiple asset objects based on the vector elements in the object vectors of any multiple asset objects; and use the cosine similarity as the asset similarity between any multiple asset objects.
[0143] It should be noted that the specific implementation of each unit in this embodiment can refer to the corresponding content in the previous text and will not be described in detail here.
[0144] refer to Figure 8 , is a schematic diagram of the structure of an electronic device provided in Embodiment 3 of the present application, and the electronic device may be an electronic device capable of data processing, such as a computer or a server, etc. The technical solution in this embodiment is mainly used to improve the accuracy of asset similarity obtained based on vectors through vectorized representation of asset objects.
[0145] Specifically, the electronic device in this embodiment may include the following structure:
[0146] The memory 801 is used to store the application program and the data generated by the application program;
[0147] Processor 802 is used to execute the application program to achieve: obtaining an asset object set, the asset object set includes multiple asset objects, the asset objects have object data, the asset objects are represented by object vectors, the object vectors include multiple vector elements, and the vector elements have initial element values; obtaining an object similarity matrix corresponding to the asset object set according to the object data of the asset objects, the matrix elements of the object similarity matrix are the initial similarities between any two asset objects in the asset object set; processing the initial element values of the vector elements in the object vector of the asset object according to the matrix elements in the object similarity matrix to obtain optimized vector elements in the object vector of the asset object, the vector elements in the object vector are used to obtain the asset similarity between the asset object to which it belongs and other asset objects.
[0148] It can be seen from the above scheme that in an electronic device provided in the third embodiment of the present application, the object data of the asset object is used to obtain the object similarity matrix corresponding to the asset object set to which it belongs, and the matrix elements in the object similarity matrix are the initial similarities between any two asset objects in the asset object set. Based on this, according to the initial similarity in the object similarity matrix, the vector elements in the object vector representing the asset object are optimized, thereby representing the asset object with the object vector composed of the optimized vector elements, and then the asset similarity between the asset objects can be obtained based on the object vector of the asset object. It can be seen that in this embodiment, after the initial similarity between all any two asset objects is obtained using the object data of the asset object, the vector elements of all asset objects represented by vectorization are optimized using the initial similarity, thereby, the asset object is represented by the optimized object vector, thereby avoiding the situation where the accuracy of the asset similarity obtained by relying solely on the data such as the yield of two asset objects is low, and the similarity between the asset objects is obtained through the optimized object vector, thereby achieving the purpose of improving the accuracy of the asset similarity.
[0149] The following is an example of obtaining the similarity of multiple assets such as stocks, funds, and indexes to illustrate the technical solution of this application in detail:
[0150] First of all, there are usually two types of implementation schemes for asset similarity analysis: quantitative analysis and qualitative analysis, but both schemes have some defects:
[0151] (1) Quantitative analysis method: The existing practice is generally to calculate the correlation coefficient of the yield series of two assets over a period of time. The disadvantage is that the yield series of assets often have missing values, resulting in inaccurate coefficients. In addition, when calculating the correlation coefficient between two assets as the similarity, this method only considers the yield series of the two assets themselves, ignoring the impact of the yield series of other assets related to the two assets on the correlation coefficient between the two assets, resulting in inaccurate correlation coefficient between the two assets.
[0152] (2) Qualitative analysis method: Taking stocks as an example, researchers will find similar company stocks from the perspectives of the industry to which the company belongs, the main business, etc. However, this method requires individual analysis of each company, and human energy is limited, so it is difficult to cover all stocks within a limited time.
[0153] In view of this, the technical solution of this application uses a vector to represent the asset, and the similarity between assets is represented by the cosine similarity between two vectors. The optimization goal is to minimize the similarity error between all assets, so as to optimize the similarity between assets globally, and avoid the correlation coefficient method that only considers the two asset yield sequences. In addition, the technical solution of this application is insensitive to the loss of yield when achieving similarity acquisition. The specific implementation scheme is shown in the following steps:
[0154] (1) Determine the asset set, which is the asset object set mentioned above, that is, determine which assets are to be vectorized, where the assets are the asset objects mentioned above. In the technical solution of this application, assets may include any number of all mutual funds, A-share stocks, CITIC Index, Shenwan Index and Wind Index.
[0155] (2) Collect the daily return sequence of each asset within a time window in the past. The size of this time window can be adjusted as needed, for example, one year or one month. The longer the time, the vector corresponding to the asset describes a relatively long-term relationship between assets, and the shorter the time, the shorter the relationship between assets.
[0156] (3) Construct an asset similarity matrix S, which is the object similarity matrix mentioned above. This similarity matrix is a matrix of size N*N, where N is the number of assets. Each element in the matrix is the similarity between every two assets obtained in the sample. The sample refers to the yield sequence within the time window determined in step (2). Therefore, this matrix is a symmetric matrix, that is, S[i,j] = S[j,i], i≠j;
[0157] Specifically, the calculation method of each matrix element in the matrix can refer to formula (3). ij represents the observed similarity between asset i and asset j, r i,t represents the rate of return of asset i on day t, and T is the size of the selected time window. The similarity matrix S can be constructed according to formula (3).
[0158] (4) Asset vector training. Based on the similarity matrix S constructed in step (3), the asset vector is trained, and the vector of asset i is defined as V i (Vector V iThe dimension can be selected as 100 and can be adjusted in different scenarios). The optimization target of training refers to formula (1) and formula (2) and the corresponding contents.
[0159] The training method can use the gradient descent algorithm. The algorithm is iterated until convergence to obtain the vector V of asset i. i .
[0160] It can be seen that the technical solution of the present application proposes a set of algorithms to vectorize the assets, that is, to use vectors to represent an asset. As long as there is daily price data for an asset, it can be included in the vectorization algorithm framework of the present application. Among them, this vector implies the asset's return volatility characteristics and industry sector information, etc., that is, assets with similar return volatility have similar corresponding vector distances, and assets in the same industry or the same sector also have similar corresponding vector distances. Based on this characteristic of asset vectors, it is possible to calculate and analyze the similarity between assets, cluster assets, and group similar assets together for subsequent analysis and strategy research, and the asset vector can be regarded as a series of factors as input variables of the supervised learning model.
[0161] Fig. 9 In the figure, the asset vector is visualized, that is, the high-dimensional asset vector is reduced to three-dimensional space. Each point represents an asset, and the area where points are clustered represents that these assets are very similar, so they are clustered in space, and the distance between assets in the same industry / sector or with similar volatility characteristics in space is also similar.
[0162] In summary, the technical solution of the present application uses vectorization technology to characterize and describe assets. It is an innovative method for analyzing the relationship between assets. Through empirical research, it is found that the characterization effect of asset similarity in multiple scenarios is better than the traditional quantitative method. In addition, the asset vectorization method proposed in the technical solution of the present application is a very flexible algorithm framework. In different scenarios, it is necessary to characterize and analyze the similarity between assets from different angles. Therefore, the similarity matrix S can be constructed from different data perspectives. For example, in addition to the use of yield data to construct the similarity matrix S mentioned in this application, the fundamental data or other data of the asset can also be used to construct the similarity matrix S. That is, different methods of constructing the similarity matrix S represent different perspectives for characterizing asset similarity. At the same time, the technical solution of the present application does not require manual intervention, so it can realize automatic regular updating of asset vectors, because the similarity between assets is different in different time windows. The technical solution of the present application can track the latest asset relationships in a regular updated manner.
[0163] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.
[0164] Professionals may further appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in the above description according to function. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians may use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.
[0165] The steps of the method or algorithm described in conjunction with the embodiments disclosed herein may be implemented directly using hardware, a software module executed by a processor, or a combination of the two. The software module may be placed in a random access memory (RAM), a memory, a read-only memory (ROM), an electrically programmable ROM, an electrically erasable programmable ROM, a register, a hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art.
[0166] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.
Claims
1. A data processing method, characterized in that: The method comprises: Obtain an asset object set, wherein the asset object set includes a plurality of asset objects, the asset objects have object data, the asset objects are represented by object vectors, the object vectors include a plurality of vector elements, and the vector elements have initial element values; According to the object data of the asset objects, an object similarity matrix corresponding to the asset object set is obtained, wherein the matrix elements of the object similarity matrix are the initial similarities between any two asset objects in the asset object set; According to the matrix elements in the object similarity matrix, the initial element values of the vector elements in the object vector of the asset object are processed to obtain the optimized vector elements in the object vector of the asset object, and the vector elements in the object vector are used to obtain the asset similarity between the asset object to which it belongs and other asset objects; The initial element values of the vector elements in the object vector of the asset object are processed according to the matrix elements in the object similarity matrix to obtain the optimized vector elements in the object vector of the asset object, including: Taking the initial similarity in the matrix elements in the object similarity matrix as the input of the similarity error function, and optimizing the initial element values of the vector elements in the object vector of the asset object by using the similarity error function, so as to obtain the optimized vector elements in the object vector of the asset object; The similarity error function is represented by at least the sum of similarity errors between all arbitrary two asset objects in the asset object set, wherein the similarity error between the asset objects in the similarity error function is represented by at least the vector distance between the object vectors of the corresponding asset objects and the initial similarity between the corresponding asset objects; The object data includes yield data corresponding to multiple dates; the matrix elements of the object similarity matrix are obtained in the following manner: Obtain the yield difference of the object data of the two asset objects corresponding to the matrix elements on each date; At least according to the yield difference on each of the dates, the initial similarity between the two asset objects corresponding to the matrix elements is obtained.
2. The method according to claim 1, characterized in that Taking the initial similarity of the matrix elements in the object similarity matrix as the input of the similarity error function, and using the similarity error function to optimize the initial element values of the vector elements in the object vector of the asset object, including: Using the initial similarity in the matrix elements in the object similarity matrix as the input of the similarity error function, obtaining a first output function value of the similarity error function for the initial element value of the vector element in the object vector of the asset object; According to the first output function value, adjusting the initial element value of the vector element in the object vector of the asset object; Taking the initial similarity in the matrix elements in the object similarity matrix as the input of the similarity error function, obtaining a second output function value of the similarity error function for the adjusted element value of the vector element in the object vector of the asset object; At least according to the second output function value, adjust the adjusted element value of the vector element in the object vector of the asset object, and return to execute the step of: using the initial similarity in the matrix element in the object similarity matrix as the input of the similarity error function, obtaining the second output function value of the similarity error function for the adjusted element value of the vector element in the object vector of the asset object, until the adjusted element value of the vector element in the object vector of the asset object satisfies the iterative convergence condition, so as to obtain the optimized vector element in the object vector of the asset object; The iteration convergence condition is that the change in the adjusted element value of the vector element in the object vector of the asset object is less than or equal to a change threshold.
3. The method according to claim 1, characterized in that In the similarity error between the two asset objects in the similarity error function, the initial similarity between the two asset objects is processed by taking logarithm; The similarity error between the two asset objects in the similarity error function is also represented by vector offset items corresponding to the object vectors of the two asset objects, and the vector offset items are related to the vector elements in the object vectors corresponding thereto; The similarity error between the two asset objects in the similarity error function is squared; The similarity error between the two asset objects in the similarity error function is weighted.
4. The method according to claim 1, characterized in that: At least according to the yield difference on each of the dates, obtaining the initial similarity between the two asset objects corresponding to the matrix elements, comprising: The yield difference of the object data of the two asset objects corresponding to the matrix elements on each date is processed respectively to obtain the similarity component corresponding to each date; The similarity components are accumulated according to the dates to obtain an initial similarity between the two asset objects corresponding to the matrix elements.
5. The method according to claim 4, characterized in that The yield difference of the object data of the two asset objects corresponding to the matrix elements on each date is processed respectively to obtain the similarity component corresponding to each date, including: Respectively obtain the absolute value of the yield difference between the object data of the two asset objects corresponding to the matrix elements on each date; The absolute values of the yield differences are processed respectively to obtain the similarity components corresponding to each of the dates, and the similarity components corresponding to each of the dates are values greater than or equal to 0.
6. The method according to claim 1, characterized in that The method further comprises: Obtaining vector elements in object vectors of any plurality of the asset objects in the asset object set; According to the vector elements in the object vectors of the arbitrary multiple asset objects, the cosine similarity between the object vectors of the arbitrary multiple asset objects is obtained; and the cosine similarity is used as the asset similarity between the arbitrary multiple asset objects.
7. A data processing device, characterized in that: The device comprises: An object obtaining unit, used for obtaining an asset object set, wherein the asset object set includes a plurality of asset objects, the asset objects have object data, the asset objects are represented by object vectors, the object vectors include a plurality of vector elements, and the vector elements have initial element values; A matrix obtaining unit, configured to obtain an object similarity matrix corresponding to the asset object set according to the object data of the asset object, wherein the matrix elements of the object similarity matrix are initial similarities between any two asset objects in the asset object set; a vector optimization unit, configured to process initial element values of vector elements in the object vector of the asset object according to matrix elements in the object similarity matrix to obtain optimized vector elements in the object vector of the asset object, wherein the vector elements in the object vector are used to obtain asset similarities between the asset object to which the vector elements belong and other asset objects; The vector optimization unit is used to: use the initial similarity in the matrix elements in the object similarity matrix as the input of the similarity error function, and optimize the initial element values of the vector elements in the object vector of the asset object using the similarity error function to obtain the optimized vector elements in the object vector of the asset object; wherein the similarity error function is represented by at least the sum of similarity errors between all arbitrary two asset objects in the asset object set, wherein the similarity error between the asset objects in the similarity error function is represented by at least the vector distance between the object vectors of the corresponding asset objects and the initial similarity between the corresponding asset objects; The object data includes yield data corresponding to a plurality of dates; the matrix obtaining unit obtains the matrix elements of the object similarity matrix in the following manner: Obtain the yield difference between the object data of the two asset objects corresponding to the matrix elements on each of the dates; and obtain the initial similarity between the two asset objects corresponding to the matrix elements at least based on the yield difference on each of the dates.
8. An electronic device, characterized in that: include: A memory, used to store applications and data generated by the operation of the applications; A processor is used to execute the application program to achieve: obtaining an asset object set, wherein the asset object set includes a plurality of asset objects, the asset objects have object data, the asset objects are represented by object vectors, the object vectors include a plurality of vector elements, and the vector elements have initial element values; According to the object data of the asset objects, an object similarity matrix corresponding to the asset object set is obtained, wherein the matrix elements of the object similarity matrix are the initial similarities between any two asset objects in the asset object set; According to the matrix elements in the object similarity matrix, the initial element values of the vector elements in the object vector of the asset object are processed to obtain the optimized vector elements in the object vector of the asset object, and the vector elements in the object vector are used to obtain the asset similarity between the asset object to which it belongs and other asset objects; The initial element values of the vector elements in the object vector of the asset object are processed according to the matrix elements in the object similarity matrix to obtain the optimized vector elements in the object vector of the asset object, including: Taking the initial similarity in the matrix elements in the object similarity matrix as the input of the similarity error function, and optimizing the initial element values of the vector elements in the object vector of the asset object by using the similarity error function, so as to obtain the optimized vector elements in the object vector of the asset object; The similarity error function is represented by at least the sum of similarity errors between all arbitrary two asset objects in the asset object set, wherein the similarity error between the asset objects in the similarity error function is represented by at least the vector distance between the object vectors of the corresponding asset objects and the initial similarity between the corresponding asset objects; The object data includes yield data corresponding to multiple dates; the matrix elements of the object similarity matrix are obtained in the following manner: Obtain the yield difference of the object data of the two asset objects corresponding to the matrix elements on each date; At least according to the yield difference on each of the dates, the initial similarity between the two asset objects corresponding to the matrix elements is obtained.
Citation Information
Patent Citations
Financial investment analysis method based on quantum clustering algorithm, storage medium and terminal
CN113674097A
Method and apparatus for calculating link-based similarity using reachability
KR1020130103079A
