Interactive method and method for building an investment model

Abstract

The embodiment of the specification provides a method for mining factors, a method for constructing an investment model and an interaction method. The method for mining factors comprises the following steps: selecting a candidate element as an exploration node based on a preset node selection strategy to expand an exploration tree. The preset node selection strategy comprises the following steps: selecting according to return information when the candidate element is used as the exploration node. The candidate element comprises a factor generation operator and an initial feature. A factor tree is obtained from the exploration tree. The factor tree is a complete subtree extended from a child node under a root node of the exploration tree. An initial feature of a leaf node in the factor tree is used to evaluate a factor generation operator of a non-leaf node in the factor tree. A factor value of the factor tree is determined. The factor value and a return information evaluation strategy are used to evaluate return information of the factor tree. The return information is propagated to a root node direction of the exploration tree to update return information of exploration nodes on a propagation path. According to the return information of each factor tree, a factor tree satisfying a preset return condition is determined.

Description

Technical Field

[0001] The embodiments in this specification relate to the field of computer technology, and in particular to an interactive method. Background Technology

[0002] Multi-factor models are commonly used machine learning models. In constructing a multi-factor model, several factors are needed to predict the results, and good predictions rely heavily on a large number of high-quality factors. A comprehensive, powerful, and extensive factor library is the source of excellent predictive ability in machine learning models. Currently, experts observe patterns and mine factors based on experience, which is not only inefficient but also results in factors that may not accurately explain the results, leading to inaccurate predictions in the constructed model. Therefore, how to efficiently and quickly mine factors with stable and accurate predictive performance has become a challenge in constructing machine learning models. Summary of the Invention

[0003] In view of this, embodiments of this specification provide a method for mining factors. One or more embodiments of this specification also relate to an interactive method, a computing device, a computer-readable storage medium, and a computer program to address technical deficiencies in the prior art.

[0004] According to a first aspect of the embodiments of this specification, a method for mining factors is provided, comprising: selecting candidate elements as exploration nodes based on a preset node selection strategy, and expanding an exploration tree, wherein the preset node selection strategy includes: selecting based on the reward information when the candidate elements are exploration nodes, wherein the candidate elements include: factor generation operators and initial features; obtaining a factor tree from the exploration tree, wherein the factor tree is a complete subtree extended from the child exploration nodes under the root node of the exploration tree, wherein the factor generation operators in the complete subtree are non-leaf nodes and the initial features are leaf nodes; evaluating the factor generation operators of the non-leaf nodes in the factor tree using the initial features of the leaf nodes in the factor tree to determine the factor values ​​corresponding to the factor tree; evaluating the reward information of the factor tree using the factor values ​​of the factor tree and a preset reward information evaluation strategy, and propagating the reward information of the factor tree from the leaf nodes of the factor tree towards the root node of the exploration tree to update the reward information of the exploration nodes on the propagation path; and determining factor trees that satisfy preset reward conditions based on the reward information of each factor tree.

[0005] According to a second aspect of the embodiments of this specification, another method for mining factors is provided, comprising: selecting candidate elements as exploration nodes based on a preset node selection strategy, and expanding an exploration tree, wherein the preset node selection strategy includes: selecting based on the reward information when the candidate elements are exploration nodes, wherein the candidate elements include: alpha factor generation operators and quantity-price data; obtaining an alpha factor tree from the exploration tree, wherein the alpha factor tree is a complete subtree extended from the child exploration nodes under the root node of the exploration tree, wherein the alpha factor generation operators in the complete subtree are non-leaf nodes and the quantity-price data are leaf nodes; evaluating the alpha factor generation operators of the non-leaf nodes in the alpha factor tree using the quantity-price data of the leaf nodes in the alpha factor tree to determine the factor value corresponding to the alpha factor tree; evaluating the reward information of the alpha factor tree using the factor value of the alpha factor tree and a preset reward information evaluation strategy, and propagating the reward information of the alpha factor tree from the leaf nodes of the alpha factor tree to the root node of the exploration tree to update the reward information of the exploration nodes on the propagation path; and determining alpha factor trees that satisfy preset reward conditions based on the reward information of each alpha factor tree.

[0006] According to a third aspect of the embodiments of this specification, a method for constructing an investment model is provided, comprising: predefining candidate elements to obtain a set of candidate elements, the candidate elements including: an alpha factor generation operator and price and volume data; performing factor mining using the factor mining method as described in any embodiment of this specification and the set of candidate elements to obtain an alpha factor tree that satisfies preset return conditions; and constructing an investment model based on the alpha factor tree that satisfies the preset return conditions.

[0007] According to a fourth aspect of the embodiments of this specification, an interactive method is provided, comprising: providing a candidate element selection interface, the candidate element selection interface including a scene selection component and a candidate element display component; in response to a user selecting any preset scene using the scene selection component in the candidate element selection interface, displaying a list of factor generation operators corresponding to the preset scene through the candidate element display component, wherein the list of factor generation operators corresponding to different preset scenes is different; obtaining the factor generation operators selected by the user from the list of factor generation operators and obtaining the initial features input by the user as a set of candidate elements; and performing factor mining using the factor mining method as described in any embodiment of this specification and the set of candidate elements to mine a factor tree that satisfies preset reward conditions.

[0008] According to a fifth aspect of the embodiments of this specification, a computing device is provided, comprising: a memory and a processor; the memory is configured to store computer-executable instructions, and the processor is configured to execute the computer-executable instructions, wherein the computer-executable instructions, when executed by the processor, implement the steps of the method for mining factors as described in any embodiment of this specification, or implement the steps of the method for constructing an investment model as described in any embodiment of this specification.

[0009] According to a sixth aspect of the embodiments of this specification, a computer-readable storage medium is provided that stores computer-executable instructions, which, when executed by a processor, implement the steps of the method for mining factors as described in any embodiment of this specification, or the steps of the method for constructing an investment model as described in any embodiment of this specification.

[0010] According to a seventh aspect of the embodiments of this specification, a computer program is provided, wherein when the computer program is executed in a computer, it causes the computer to perform the steps of the method for mining factors described above, or the steps of the method for constructing an investment model described above.

[0011] One embodiment of this specification implements a method for mining factors. This method selects candidate elements as exploration nodes based on a preset node selection strategy, expanding the exploration tree. The preset node selection strategy includes: selecting based on the reward information when candidate elements are used as exploration nodes; the candidate elements include: factor generation operators and initial features; obtaining a factor tree from the exploration tree, wherein the factor tree is a complete subtree extended from the child exploration nodes under the root node of the exploration tree, with the factor generation operators as non-leaf nodes and the initial features as leaf nodes in the complete subtree; evaluating the factor generation operators of the non-leaf nodes in the factor tree using the initial features of the leaf nodes in the factor tree to determine the factor values ​​corresponding to the factor tree; evaluating the reward information of the factor tree using the factor values ​​and a preset reward information evaluation strategy, and propagating the reward information of the factor tree from the leaf nodes of the factor tree towards the root node of the exploration tree to update the reward information of the exploration nodes along the propagation path; and determining factor trees that satisfy preset reward conditions based on the reward information of each factor tree. As can be seen, this method expands the exploration tree through a heuristic exploration algorithm, continuously selecting candidate elements and expanding exploration nodes. During the expansion of the exploration tree, the child exploration nodes under the root node of the exploration tree grow into factor trees. The exploration process of factor trees is guided by the exploration tree, and the exploration tree and factor trees grow simultaneously, avoiding the problem of insufficient diversity. Applying the exploration tree to factor mining based on factor generation operator expressions improves search efficiency and diversity, and the mined factors have more stable and accurate prediction effects. Attached Figure Description

[0012] Figure 1 This is a schematic diagram illustrating an application scenario of a factor mining method provided in one embodiment of this specification in mining the alpha factor of an investment model;

[0013] Figure 2 This is a flowchart illustrating a method for mining factors according to one embodiment of this specification;

[0014] Figure 3 This is a schematic diagram of node information provided in one embodiment of this specification;

[0015] Figure 4 This is a schematic diagram illustrating constraint information provided in one embodiment of this specification;

[0016] Figure 5 This is a schematic diagram of operator information for a factor generation operator provided in one embodiment of this specification;

[0017] Figure 6 This is a schematic diagram of node information for a factor tree node provided in one embodiment of this specification;

[0018] Figure 7 This is a schematic diagram of the exploration node growth factor tree of a Monte Carlo tree provided in one embodiment of this specification;

[0019] Figure 8 This is a schematic diagram of the factor tree structure expansion provided in one embodiment of this specification;

[0020] Figure 9 This is a schematic diagram of the device for mining factors provided in one embodiment of this specification;

[0021] Figure 10 This is a flowchart illustrating a method for constructing an investment model according to one embodiment of this specification;

[0022] Figure 11 This is a structural diagram of an apparatus for constructing an investment model according to one embodiment of this specification;

[0023] Figure 12 This is a flowchart of an interaction method provided in one embodiment of this specification;

[0024] Figure 13 This is a schematic diagram of the interface interaction of an embodiment of the interaction method provided in this specification;

[0025] Figure 14 This is a structural diagram of an interactive device provided in one embodiment of this specification;

[0026] Figure 15 This is a structural block diagram of a computing device provided in one embodiment of this specification. Detailed Implementation

[0027] Many specific details are set forth in the following description to provide a full understanding of this specification. However, this specification can be implemented in many other ways than those described herein, and those skilled in the art can make similar extensions without departing from the spirit of this specification. Therefore, this specification is not limited to the specific implementations disclosed below.

[0028] The terminology used in one or more embodiments of this specification is for the purpose of describing particular embodiments only and is not intended to be limiting of the one or more embodiments of this specification. The singular forms “a,” “described,” and “the” as used in one or more embodiments of this specification and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used in one or more embodiments of this specification refers to and includes any or all possible combinations of one or more associated listed items.

[0029] It should be understood that although the terms first, second, etc., may be used to describe various information in one or more embodiments of this specification, such information should not be limited to these terms. These terms are only used to distinguish information of the same type from one another. For example, first may also be referred to as second without departing from the scope of one or more embodiments of this specification, and similarly, second may also be referred to as first. Depending on the context, the word "if" as used herein may be interpreted as "when," "when," or "in response to a determination."

[0030] First, the terms and concepts used in one or more embodiments of this specification will be explained.

[0031] An exploration tree is an exploration algorithm that uses a tree data structure to balance exploration and utilization, and finds exploration nodes within the exploration space that offer satisfactory rewards. An example is an exploration tree called a Monte Carlo tree.

[0032] Monte Carlo trees are a heuristic exploration algorithm that builds an exploration tree through a process of selection, expansion, simulation, and backpropagation, and then uses this exploration tree to solve a given problem in the target space.

[0033] Alpha factor: A characteristic variable used in multi-factor models to explain the excess returns of individual stocks.

[0034] To address the issues of low efficiency, inaccurate interpretation of results, and inaccurate predictions in existing factor mining methods, this specification provides an exploration algorithm using an exploration tree. This algorithm expands the exploration tree by continuously selecting candidate elements and expanding exploration nodes. During the expansion, a factor tree grows from the child exploration nodes under the root node of the exploration tree. The exploration process of the factor tree is guided by the exploration tree, and the exploration tree and factor tree grow simultaneously, avoiding the problem of insufficient diversity. Applying the exploration tree to factor mining based on factor generation operator expressions improves search efficiency and diversity, and the mined factors have more stable and accurate predictive effects.

[0035] Specifically, this specification provides methods for factor mining, methods for constructing investment models, and interactive methods. This specification also relates to apparatus for factor mining, apparatus for constructing investment models, interactive apparatus, computing devices, and computer-readable storage media, which will be described in detail in the following embodiments.

[0036] See Figure 1 , Figure 1 The diagram illustrates an application scenario of the method for mining factors according to an embodiment of this specification in mining the alpha factor of an investment model.

[0037] like Figure 1 As shown, in this application scenario, the investment model building side can submit a mining request to the alpha factor mining side. The mining request can carry candidate elements, including: alpha factor generation operators and price and volume data. The alpha factor mining side responds to the mining request by executing the factor mining method, specifically including:

[0038] Obtain the alpha factor generation operator and quantity-price data;

[0039] Candidate elements are selected as exploration nodes based on a preset node selection strategy to expand the exploration tree. The preset node selection strategy includes: selecting based on the reward information when candidate elements are used as exploration nodes.

[0040] An alpha factor tree is obtained from the exploration tree, wherein the alpha factor tree is a complete subtree extended from the child exploration node under the root node of the exploration tree, and the alpha factor generating operator is used as a non-leaf node and the quantity and price data is used as a leaf node in the complete subtree;

[0041] The alpha factor generation operator of the non-leaf nodes in the alpha factor tree is evaluated using the quantity and price data of the leaf nodes in the alpha factor tree to determine the factor value corresponding to the alpha factor tree.

[0042] Using the factor values ​​of the alpha factor tree and a preset reward information evaluation strategy, the reward information of the alpha factor tree is evaluated, and the reward information of the alpha factor tree is propagated from the leaf node of the alpha factor tree to the root node of the exploration tree to update the reward information of the exploration nodes on the propagation path.

[0043] Based on the return information of each alpha factor tree, determine the alpha factor trees that meet the preset return conditions.

[0044] Furthermore, the alpha factor mining side can also return the mined alpha factor trees that meet preset return conditions to the investment model building side. The investment model building side obtains the alpha factor trees returned by the alpha factor mining side and puts them into a factor library. When the alpha factors in the factor library are sufficient to train the investment model, the investment model building side uses the alpha factors in the factor library to train the investment model, obtaining a trained investment model. In addition, after obtaining the trained investment model, the investment model building side can also provide the investment model to the user side so that the user side can use the investment model to predict returns.

[0045] In this system, connections are established between the user side and the investment model building side, and between the investment model building side and the alpha factor mining side, via networks. These networks can include various connection types, such as wired or wireless communication links or fiber optic cables.

[0046] The user side can manifest as a browser, an app (application), or a web application such as an H5 (HyperText Markup Language 5) application, a lightweight application (also known as a mini-program), or a cloud application. The user side can be based on the SDK (Software Development Kit) provided by the investment model building side, such as development based on the RTC (Real-Time Communication) SDK. The user side can be deployed in electronic devices, requiring the device to run or certain apps within the device to function. Electronic devices can have displays and support information browsing, such as personal mobile terminals like smartphones, tablets, and personal computers. Various other types of applications can also be configured on electronic devices, such as human-computer interaction applications, model training applications, text processing applications, web browser applications, shopping applications, search applications, instant messaging tools, email object supply terminals, and social platform software.

[0047] The investment model building side and the alpha factor mining side can run on servers that provide various services, such as information services, background training services for investment models, and services that process requests from users.

[0048] Servers used for deploying investment model building and alpha factor mining can be implemented as a distributed server cluster consisting of multiple servers, or as a single server. Servers can also be servers in a distributed system, or servers integrated with blockchain. Servers can also be cloud servers providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms, or intelligent cloud computing servers or intelligent cloud hosts with artificial intelligence technology.

[0049] It should be noted that, Figure 1 The scenarios illustrated are merely illustrative of the methods for factor mining and investment model construction provided in the embodiments of this specification, and do not constitute a limitation on the methods provided in the embodiments of this specification. The methods for factor mining and investment model construction can be executed by a server or a terminal, or jointly by a terminal and a server. These methods can be applied to factor mining and model construction in machine learning models for any predictive purpose. For example, transaction behavior prediction models in online trading scenarios, content prediction models in content push scenarios, and so on.

[0050] See Figure 2 , Figure 2 A flowchart illustrating a method for mining factors according to one embodiment of this specification is shown. Figure 2 As shown, the method for mining factors specifically includes the following steps.

[0051] Step 202: Select candidate elements as exploration nodes based on a preset node selection strategy to expand the exploration tree. The preset node selection strategy includes: selecting based on the reward information when candidate elements are used as exploration nodes. The candidate elements include: factor generation operators and initial features.

[0052] The factor generation operator refers to the factor expression used to calculate factor values, which can also be understood as the feature variable used to explain the prediction results in a machine learning model. The goal of machine learning is to use the feature variables of data to predict the target. Factor generation operators can be represented by operators such as ADD, SUM, MAX, MIN, STD, EXP, TSMEAN, and GTE. For example, in investment model applications, the factor generation operator can be represented by various alpha factor generation operators used to explain investment returns.

[0053] ADD stands for arithmetic addition, where ADD(a,b) = a + b;

[0054] SUM is a moving summation, and SUM(a,w) is a summation of the sequence with w as the sliding window.

[0055] MAX is used to find the maximum value; MAX(a,b) means taking the maximum value between a and b.

[0056] MIN is used to find the minimum value; MIN(a,b) means taking the minimum value between a and b.

[0057] STD is used to calculate the rolling standard deviation, and STD(a,w) is used to calculate the standard deviation of the sequence with w as the window.

[0058] EXP is for exponentiation, and EXP(a,x) is for raising each element in the sequence to the power of x.

[0059] TSMEAN is a moving average, and TSMEAN(a,w) is the average of the series with w as the sliding window.

[0060] GTE stands for logical operation. GET(a,b) means comparing each element in a and b to see if it is greater than or equal to a.

[0061] The initial features are feature values ​​with lower dimensionality compared to the factor generation operator. They can also be understood as some basic features used in machine learning models to explain prediction results. For example, the initial features can be the data sequences involved in the aforementioned factor generation operator. As another example, in investment model applications, the initial features can be represented by price and volume data recorded by investment institutions over a period of time, which can be expressed as a sequence of stock opening price, highest price, lowest price, and closing price.

[0062] The preset node selection strategy can be flexibly configured according to the application scenario. By setting the preset node selection strategy, nodes are selected based on the reward information when candidate elements are used as exploration nodes, expanding the exploration nodes. Then, through backpropagation of evaluation and reward, the exploration tree is expanded, avoiding exhaustive calculation of the reward for each candidate element at every level of the exploration tree. When the exploration space is large and the value of all subtrees cannot be calculated, it can simultaneously take into account exploration and utilization, avoiding getting trapped in local optima.

[0063] For example, the preset node selection strategy can employ the classic UCB (Upper Confidence Bounds) algorithm. This algorithm prioritizes unexplored nodes when selecting exploration nodes. If all nodes have been explored, selection is based on the score determined by the reward. The score is not only determined by the reward of the exploration node but is also negatively correlated with the number of times the node has been explored. In other words, a higher average score makes a node more likely to be selected, while a node selected more often is less likely to be selected next time. Therefore, by configuring exploration using upper confidence intervals and utilizing different weights, a more insightful exploration effect than random or other strategies can be achieved. For instance, in the UCB algorithm, weights r and c are used to adjust the exploration intensity during the search process. When r approaches 0, it resembles random exploration; when c approaches 0, it resembles greedy exploration. The score can also be called the UCB value.

[0064] The factor tree reward information is obtained by evaluating the factor values ​​obtained from the exploration tree and is used to measure the quality of the factor tree.

[0065] A factor tree is a complete subtree derived from an exploration tree, extending from the child exploration nodes under the root node of the exploration tree. A complete subtree is defined as a subtree containing intermediate nodes and leaf nodes, where the number of child nodes of all non-leaf nodes satisfies the requirement for the number of input values ​​of the non-leaf nodes.

[0066] The reward information of an exploratory node is the total reward information propagated along all propagation paths passing through that exploratory node; that is, the total reward is the sum of the backpropagation rewards from all leaf nodes under that exploratory node. It should be noted that for a child exploratory node of the root node of the exploratory tree, that child exploratory node may grow into several subtrees, some of which are complete subtrees, and some are incomplete subtrees. For an exploratory node contained in an incomplete subtree, since it may also be an exploratory node in a complete subtree, it still has significance as an intermediate state in the exploratory tree, used to continue expanding exploratory nodes in order to expand into other complete subtrees.

[0067] It's important to note that the evaluation strategy used to assess returns can be flexibly configured according to the specific application scenario. For example, in the investment field, return information can be represented as the rate of return, and the preset return information evaluation strategy can be correspondingly represented as an evaluation strategy for assessing the rate of return. For instance, for a stock, the root node of the factor tree is ADD, and ADD has two child nodes, open and close. Corresponding to a 100-day statistical period, open and close are both arrays with 100 elements each. The sum of open and close is the value of ADD, represented as a list of 100 elements. A correlation coefficient can be calculated between this list and the future rate of return to represent the rate of return, or other indicators used in the investment field to measure returns can be used as return information.

[0068] Step 204: Obtain the factor tree from the exploration tree, wherein the factor tree is a complete subtree extended from the child exploration node under the root node of the exploration tree, and the factor generation operator in the complete subtree is a non-leaf node and the initial feature is a leaf node.

[0069] It is understandable that the root node of the factor tree is a child exploration node of the root node of the exploration tree, which is also the second-level exploration node of the Monte Carlo tree. Therefore, as the exploration tree expands, different factor trees are obtained with each second-level exploration node as its root node.

[0070] The reason why the method provided in the embodiments of this specification requires the factor tree to be a complete tree is that only when the number of child nodes of each non-leaf node can meet the input value requirements of that non-leaf node can each non-leaf node perform feature evaluation, thereby completing the subsequent evaluation of the return information.

[0071] Step 206: Calculate the factor generation operators of the non-leaf nodes in the factor tree using the initial features of the leaf nodes in the factor tree, and determine the factor values ​​corresponding to the factor tree.

[0072] For example, suppose the root node of the factor tree is the factor generating operator ADD. The ADD operator requires two input values. Based on this requirement, the root node has two child nodes: one child node contains the price data `open` (including 100 days' opening prices), and the other child node contains the price data `close` (including 100 days' closing prices). Therefore, the value of the root node ADD is equal to the sum of the 100 days' opening prices and the 100 days' closing prices; that is, the factor value of this factor tree is the value of ADD.

[0073] Step 208: Using the factor values ​​of the factor tree and the preset reward information evaluation strategy, evaluate the reward information of the factor tree, and propagate the reward information of the factor tree from the leaf nodes of the factor tree to the root node of the exploration tree to update the reward information of the exploration nodes on the propagation path.

[0074] Taking the Monte Carlo tree as an example, the process of expanding the Monte Carlo tree involves four steps: selection, expansion, simulation, and backpropagation. Selection: This involves finding the most worthwhile exploration node based on the UCB algorithm. A common strategy is to first select unexplored candidate elements; if all have been explored, then, based on the UCB algorithm, select the candidate element with the highest score from the expanded exploration nodes. Expansion: After the selection, the corresponding exploration node is expanded within the Monte Carlo tree. Simulation: This involves growing the factor tree of the expanded exploration node, evaluating the factor values, and calculating the reward information. Backpropagation: This involves propagating the reward information of the factor tree back along the growth path from the leaf node to all previous parent nodes until the root node of the exploration tree. This updates the reward information and visit count of the exploration nodes along the propagation path, facilitating the subsequent calculation of the UCB value, i.e., the score.

[0075] Step 210: Based on the return information of each factor tree, determine the factor trees that meet the preset return conditions.

[0076] The preset return conditions are used to filter factor trees that meet the model's requirements based on return information. Specifically, they can be set according to the criteria for factor trees required by the model in the actual application scenario. For example, the preset return conditions can be expressed as a standard return threshold range corresponding to one or more return information types. When the return information of a certain factor tree falls within the corresponding standard return threshold range, it is determined to be a factor tree that meets the preset return conditions. For example, the standard return threshold range can be an range greater than a standard return threshold.

[0077] This method is based on the heuristic exploration algorithm of the exploration tree. It expands the exploration tree by continuously selecting candidate elements and expanding exploration nodes. During the expansion of the exploration tree, the exploration nodes grow into factor trees. The exploration process of the factor trees is guided by the exploration tree. The exploration tree and factor trees grow simultaneously, avoiding the population aging problem caused by insufficient diversity in genetic algorithms with iteration. Applying the exploration tree to factor mining based on factor generation operator expressions improves search efficiency and diversity, and results in better mining effects.

[0078] In one or more embodiments of this specification, the node information of the exploration nodes in the exploration tree includes: the report information of the exploration nodes. Correspondingly, the selection based on the report information when a candidate element is used as an exploration node includes: when selecting a child exploration node for any exploration node, if there are remaining candidate elements among the candidate elements that have not yet been selected, then a candidate element is selected from the remaining candidate elements as a child exploration node of the any exploration node; if there are no remaining candidate elements among the candidate elements that have not yet been selected, then a candidate element is selected from the expanded exploration nodes according to the report information recorded in the node information of the expanded exploration nodes in the exploration tree, as a child exploration node of the any exploration node.

[0079] For example, each time the reward information of the factor tree is evaluated and propagated along the growth path of the factor tree from the leaf node to the root node of the exploration tree, the reward information can be recorded in the node information of the exploration nodes traversed along the propagation path. Since the reward information recorded in the node information of the exploration node is the total reward of all reward information propagated along the propagation path of all the exploration nodes passing through that exploration node, when a new factor tree grows as the exploration tree expands, the reward information of the new factor tree is correspondingly added to the total reward of the exploration nodes traversed during the backpropagation of the reward information of the new factor tree.

[0080] Taking child explorer node A under the root node of the exploration tree as an example: Assume that explorer node A expands into child explorer nodes B and C; as the exploration tree further expands, a factor tree AB1 grows from explorer node A through explorer node B, and a factor tree AC1 grows from explorer node A through explorer node C; the reward information of factor tree AB1 is V(AB1), and the reward information of factor tree AC1 is V(AC1); the reward value V(AB1) propagated backward from the leaf node of factor tree AB1 and the reward value V(AC1) propagated backward from the leaf node of factor tree AC1 both pass through explorer node A, thus updating the total reward in the node information of explorer node A to AB1 + AC1.

[0081] In the above embodiments, the propagated reward information is recorded in the node information of the exploration node through backpropagation. This makes it easier to obtain the reward information from the node information when selecting an exploration node, so as to use the preset node selection strategy to select the exploration node, ensuring the optimization of the exploration tree and the compromise of the exploration process.

[0082] In one or more embodiments of this specification, the node information of the exploration nodes in the exploration tree includes: node effectiveness information, which indicates whether the exploration node is effective. The preset node selection strategy further includes: selecting based on the node effectiveness information when a candidate element is used as an exploration node. Accordingly, the method further includes: evaluating the effectiveness of the factor generation operator corresponding to the exploration node using a preset node effectiveness evaluation strategy to obtain the effectiveness evaluation information of the exploration node; and propagating the node effectiveness information of the exploration node from the exploration node towards the root node of the exploration tree. The step of selecting based on the node effectiveness information when a candidate element is used as an exploration node includes: when selecting a sub-exploration node for any exploration node in the exploration tree, if a candidate element is selected from the expanded exploration nodes of the exploration tree, then a candidate element with effective nodes is selected based on the node effectiveness information recorded in the node record area of ​​the expanded exploration node.

[0083] The preset node effectiveness evaluation strategy can be set according to the actual application scenario, that is, the invalidity criteria can be flexibly set. For example, the preset node effectiveness evaluation strategy may include: for the value of the factor generation operator corresponding to the exploration node, if the value is a constant or empty, then the corresponding exploration node is confirmed as invalid.

[0084] It is understandable that when the node validity information of an explorer node propagates from that explorer node toward the root node of the explorer tree, it first passes through the parent explorer node. Only when all child explorer nodes of the parent explorer node are invalid will the node validity information of the parent explorer node be updated to invalid; otherwise, the node validity information of the parent explorer node is valid.

[0085] Taking exploration node A as an example, suppose exploration node A expands into child exploration node B. Evaluating the effectiveness of the factor generation operator of exploration node B reveals that its node effectiveness information is invalid. If exploration node A also has a child exploration node C besides B, and the node effectiveness information of child exploration node C is valid, then after the propagation of node effectiveness information, the node effectiveness information of exploration node A remains valid. If the node effectiveness information of child exploration node C is also invalid, then after the propagation of node effectiveness information, the node effectiveness information of exploration node A becomes invalid. The node effectiveness information of exploration node A continues to propagate towards the root node of the exploration tree, which will not be elaborated further here.

[0086] In the above embodiments, by recording node validity information in the node information of the exploration node and propagating it in reverse, it is ensured that nodes that have been confirmed as invalid will not be selected again when the exploration node is selected next time, thus ensuring that "the same mistake will not be made twice" and effectively improving the efficiency of expanding the exploration tree.

[0087] In one or more embodiments of this specification, the node information of the exploration node in the exploration tree includes: exploration completeness information, which indicates whether the exploration node has completed exploration. The preset node selection strategy further includes: selecting based on the exploration completeness information when a candidate element is an exploration node. The method further includes: determining whether the factor tree corresponding to all leaf nodes extended from the exploration node has been evaluated; if so, determining that the exploration node has completed exploration, and updating the exploration completeness information in the node information of the exploration node accordingly; propagating the exploration completeness information of the exploration node from the exploration node to the root node of the exploration tree. Accordingly, the selection based on the exploration completeness information when a candidate element is an exploration node includes: when selecting a child exploration node for any exploration node of the exploration tree, if a candidate element is selected from the exploration nodes that have been extended in the exploration tree, then selecting the candidate element that has not yet completed complete exploration based on the exploration completeness information recorded in the node record area of ​​the extended exploration node.

[0088] It is understandable that the exploration completeness information indicates that the exploration node has been fully explored and therefore does not need to be explored further. Consequently, it will not be selected as an exploration node again during the subsequent construction of the exploration tree. Furthermore, exploration completeness information can help determine the overall exploration progress of the exploration tree and what proportion of the entire space has been fully explored. Therefore, the exploration completeness information of each exploration node in the exploration tree can be used as reference information for deciding whether to continue constructing the exploration tree.

[0089] For example, the exploration completeness information of an exploration node can include "yes" or "no". "Yes" indicates that the exploration node has been fully explored, while "no" indicates that the exploration node has not been fully explored. According to the above embodiment, for exploration nodes that have been fully explored, the exploration completeness information is displayed as "yes," and the exploration node will not be selected again, avoiding re-evaluation and improving exploration efficiency.

[0090] In addition, the node information of the exploration node can also be used to record tree integrity information, which indicates whether the exploration node has grown a complete subtree containing leaf nodes. For the complete subtree corresponding to the exploration node with "yes" tree integrity information, since it is possible to deduce the value of its corresponding factor generating operator from the leaf nodes upwards, recording the tree integrity information facilitates subsequent analysis. For example, only if it is a complete subtree is it eligible to be used as a factor tree; if it is incomplete, it cannot be evaluated and is therefore ineligible to be used as a factor tree.

[0091] The following is an illustrative description of an implementation method that incorporates node information from the above-described embodiments. For example... Figure 3The diagram shown illustrates the node information. The node information for each exploration node in the exploration tree can include the following items:

[0092] The node name is the identifier of the node. For example, it can be named according to the name of the corresponding candidate element, such as ADD, SUM, etc.

[0093] The child node list is a list of all direct child nodes that the current node has expanded out of;

[0094] The parent node is the direct parent node of this node;

[0095] The number of visits is the total number of paths that pass through the explored node in the exploration tree;

[0096] Total reward is the sum of the reward information of all backpropagation paths that pass through the exploration node in the exploration tree;

[0097] The execution function is the function called when the factor generation operator of the exploration node is evaluated;

[0098] The depth of the corresponding factor tree is the depth from the root node of the factor tree to the exploration node by expanding the depth-first tree.

[0099] Whether it is a complete tree is equivalent to the tree integrity information mentioned above, indicating whether the exploration node has grown into a complete factor tree containing leaf nodes;

[0100] Whether a node is bad is equivalent to the node validity information mentioned above, indicating whether the exploration node is valid;

[0101] Whether it has been fully explored is equivalent to the exploration completeness information mentioned above, indicating whether the exploration node and all exploration nodes under it have been fully explored.

[0102] The method for selecting and expanding exploration nodes in the exploration tree provided in the embodiments of this specification can be set according to the needs of actual application scenarios. For example, in one or more embodiments of this specification, the selection based on the reward information when a candidate element is used as an exploration node includes: calculating the score corresponding to the reward information when a candidate element is used as an exploration node based on the upper confidence interval algorithm, and selecting exploration tree nodes based on the score. Of course, depending on the needs of actual application scenarios, random exploration or greedy exploration methods can also be used for selection, which will not be elaborated here.

[0103] For example, in one or more embodiments of this specification, expanding the exploration tree includes expanding the exploration tree in a depth-first manner. Because the depth-first approach allows for timely determination of the exploration depth, it enables decisions on whether to continue exploring based on the depth, preventing overfitting and resulting in high exploration efficiency. Of course, depending on the specific application scenario, a breadth-first approach can also be used to expand the nodes, as long as it can generate operator expressions for the factor tree from the leaf nodes to the root node; this will not be elaborated further here.

[0104] In one or more embodiments of this specification, constraint information on the input values ​​is added to the factor generation operator to constrain the growth process of the factor tree, thereby enabling the factor tree to grow in a manner required by the actual application scenario. Specifically, the operator information of the factor generation operator includes constraint information on the input values. The preset node selection strategy further includes: when selecting a sub-exploration node, selecting candidate elements as sub-exploration nodes according to the constraint information that the output value of the sub-exploration node should satisfy the input value of the parent node of the sub-exploration node.

[0105] Optionally, the operator information of the factor generation operator may also include constraint information for the output value.

[0106] For example, the constraint information may include: constraint information on the data type of the input value at a specified input location, and / or, constraint information on the data type of the output value at a specified output location. Figure 4 The diagram illustrates the structure of constraint information. Constraint positions are numbered 1, 2, 3, etc., representing the sequence number of the input position. Additionally, if a constraint position includes an output identifier, it indicates a constraint on the output; for example, the output identifier can be represented by the letter "o". The data types of constraints can include: constant type 'const', integer type 'posint', floating-point type 'float', boolean type 'bool', and double constant type 'bothconst', etc. Furthermore, a negative constraint symbol can be added before the constraint type to indicate a constraint that is "not" the specified type. For example, adding the negative constraint symbol "!" before the constant type 'const' indicates a non-constant type. Moreover, multiple constraints can apply to the same position within a single constraint; for example, multiple constraints can be connected using a connector symbol such as '_'.

[0107] Take the factor generation operator TSMEAN as an example. TSMEAN stands for Moving Average operator. Assume the constraint information of TSMEAN is: 1!const_2posint, meaning that the first input position of TSMEAN requires the input to be a factor generation operator or initial feature whose output value (i.e., feature) does not contain a constant, and the second input position requires the input to be a factor generation operator or initial feature whose output value does not contain an integer. Correspondingly, for the node of this factor generation operator in the factor tree, when selecting the child node corresponding to the first input position, a factor generation operator or initial feature whose output value (i.e., feature) does not contain a constant will be selected, that is, a factor generation operator whose constraint information does not contain oconst will be selected; when selecting the second input, a factor generation operator or initial feature whose output value contains an integer will be selected, that is, a factor generation operator whose constraint contains oposint will be selected.

[0108] Let's take the Factor Generation Operator (GTE) as an example again. A GTE represents a comparison of two sequences. Suppose the constraint information of a GTE is !bothconst_obool, which means that at least one of the two inputs of the GTE is a non-constant, and its output is bool.

[0109] As can be seen from the above embodiments, by writing factor generation operators and corresponding constraint information, the constraint of the connection relationship between factor generation operators is realized, which enables the connection relationship between factor generation operators to conform to the convention, avoids invalid connections in the factor tree generation process, reduces invalid evaluation and assessment, and enables the exploration engine to explore in a predetermined way, thereby improving exploration efficiency.

[0110] To better represent the factor generation operator and quickly call the corresponding execution function to calculate the operator's value, in one or more embodiments of this specification, the operator information of the factor generation operator further includes: operator identification information, execution function information, and input value quantity requirement information. The operator identification information is used to identify the factor generation operator, the execution function information is the function information called when the factor generation operator is evaluated, and the input value quantity requirement information is the number of input values ​​required by the factor generation operator.

[0111] For example, such as Figure 5 The diagram shows the operator information structure of the factor generation operator, where: the operator identifier information is represented by the operator name; the execution function information can be the execution function itself or the name of the execution function, and the execution function itself can be executed directly, or the execution function can be called by the name of the execution function; the input value quantity requirement information is represented by the number of input values; and the constraint information is also known as the operator constraint.

[0112] Combining the operator information of the above factor generation operators, we can generate, as follows: Figure 6The node information of the nodes in the factor tree shown.

[0113] In conjunction with the above embodiments, in one or more embodiments of this specification, the exploration tree is an exploration tree expanded according to the Monte Carlo tree algorithm. The step of selecting candidate elements as exploration nodes based on a preset node selection strategy and expanding the exploration tree includes:

[0114] Following a depth-first approach, candidate elements are selected based on the constraint that the output value of a child exploration node should satisfy the input value of its parent node. Additionally, a score is calculated based on the upper limit confidence interval algorithm to determine the reward information corresponding to the candidate element as an exploration node. Exploration nodes are then selected based on the scores to expand the exploration tree.

[0115] For any non-leaf node in the factor tree, based on the operator identifier information, execution function information, input value quantity requirement information, and constraint information of the factor generation operator of the non-leaf node, node information of the non-leaf node is generated. The node information includes: node identifier information, execution function information, numerical quantity requirement information, parent node information, child node information, and whether it has been evaluated.

[0116] When the number of child nodes of all non-leaf nodes in the factor tree meets the requirement for the number of input values ​​of the non-leaf nodes, the factor tree is determined to have completed growth.

[0117] In the above embodiments, the exploration process is guided by a Monte Carlo tree. The second-level exploration node of the exploration tree is used as the root node of the factor tree. Child nodes are selected to grow the factor tree in a depth-first manner and according to the constraint information of the factor generation operator. The node information of the non-leaf nodes of the factor tree is generated according to the operator information of the factor generation operator. This ensures that the connection relationship between the factor generation operators conforms to the agreement, avoids invalid connections in the factor tree generation process, and improves the mining efficiency of the factor tree.

[0118] Below, in conjunction with Figure 7 The diagram illustrating the expansion process of a Monte Carlo tree serves as a schematic illustration of the factor tree growth process. The expansion process of the Monte Carlo search tree algorithm consists of selection, expansion, simulation, and backpropagation. Figure 7 As shown, the ribbon arrow pointing to the terminal node represents the forward process of selection, expansion, and simulation, while the ribbon arrow pointing to the root node represents the backward propagation process of the reward information after evaluation.

[0119] Starting from the root node of the Monte Carlo tree, at each selection, the UCB value of each child exploration node of the root node is calculated, and the child exploration node with the largest UCB value is selected. Based on the selected exploration node, the factor tree is grown according to the corresponding execution function, using a depth-first growth method.

[0120] Specifically, such as Figure 7 As shown, for the child exploration node SUB under the root node of the Monte Carlo tree, the first step is to select SUB from the candidate elements, and then expand the Monte Carlo tree to generate the child exploration node SUB of the root, with SUB as the root node of the factor tree to be grown; the second step is to select CLOSE from the candidate elements as the child exploration node of SUB, and then expand the corresponding factor tree of SUB to add the child exploration node CLOSE to SUB. Since SUB has two input values, CLOSE is the first child exploration node of SUB here; the third step is to select the candidate element DELAY, expand the factor tree, and add the child exploration node DELAY as the second child node of SUB; since the number of input values ​​of DELAY is greater than 0, the growth process is not over. In the fourth step, the candidate element CLOSE is selected, and the factor tree is expanded to add CLOSE as the first child exploration node of DELAY; in the fifth step, the initial feature 1 is selected as the second child node of DELAY, and the factor tree is expanded to generate leaf node 1. At this point, the number of child nodes of all intermediate nodes in the factor tree rooted at SUB equals the number of input values, and the growth process of the factor tree rooted at SUB is complete.

[0121] It should be noted that since the factor tree is a complete subtree of the exploration tree with the child exploration nodes of the root node of the exploration tree as its root node, expanding the factor tree is equivalent to expanding the exploration tree. In the above description of the growth process, each time a node is expanded in the factor tree, it is equivalent to expanding a node in the exploration tree; both grow simultaneously, which will not be elaborated further here.

[0122] The factor tree structure rooted at SUB, when expanded, can be represented as follows: Figure 8 The tree structure is shown. The growth path of the factor tree, i.e., the operator expression of the factor tree, can be represented as SUB(CLOSE,DELAY(CLOSE,1)). After growing the factor tree, the next step is to evaluate it. Specifically, a preset evaluator can be used to evaluate the node values ​​of each node in the factor tree and assess the reward information. This preset evaluator is a computational module used to evaluate the factor generation operators of non-leaf nodes in the factor tree using the initial features of the leaf nodes. The calculated reward information is as follows: Figure 7 The diagram shows backpropagation along the growth path. Of course, backpropagation information can also include whether a node is bad, whether the exploration is complete, etc., which will not be elaborated upon here.

[0123] As can be seen from the above exploration process, by continuously exploring nodes, the Monte Carlo tree can be used as the exploration tree for the factor tree to guide the entire exploration process, realizing a tree structure exploration scheme that allows the Monte Carlo tree and the factor tree to grow simultaneously. In the end, high-quality factor trees can be selected and stored in the factor library according to certain criteria, such as factor trees whose return information is greater than a certain threshold.

[0124] Corresponding to the above method embodiments, this specification also provides apparatus embodiments for mining factors. Figure 9 A schematic diagram of a device for mining factors according to one embodiment of this specification is shown. Figure 9 As shown, the device includes:

[0125] The exploration tree expansion module 902 is configured to select candidate elements as exploration nodes based on a preset node selection strategy to expand the exploration tree. The preset node selection strategy includes: selecting based on the reward information when candidate elements are used as exploration nodes. The candidate elements include: factor generation operators and initial features.

[0126] The factor tree acquisition module 904 is configured to obtain a factor tree from the exploration tree, wherein the factor tree is a complete subtree extended from the child exploration node under the root node of the exploration tree, and the factor generation operator in the complete subtree is a non-leaf node and the initial feature is a leaf node.

[0127] The factor tree evaluation module 906 is configured to evaluate the factor generation operators of the non-leaf nodes in the factor tree using the initial features of the leaf nodes in the factor tree when the factor tree grows into a complete tree containing leaf nodes, thereby determining the features of the factor tree.

[0128] The feedback transmission module 908 is configured to use the factor values ​​of the factor tree and a preset feedback information evaluation strategy to evaluate the feedback information of the factor tree, and propagate the feedback information of the factor tree from the leaf node of the factor tree to the root node of the exploration tree to update the feedback information of the exploration nodes on the propagation path.

[0129] The factor tree determination module 910 is configured to determine the factor trees that meet the preset return conditions based on the return information of each factor tree.

[0130] Optionally, the node information of the exploration nodes in the exploration tree includes: the report information of the exploration nodes. Correspondingly, the exploration tree expansion module 902 can be configured to, when selecting a child exploration node for any exploration node, if there are remaining unselected candidate elements among the candidate elements, select a candidate element from the remaining candidate elements as a child exploration node of any exploration node; if there are no unselected remaining candidate elements among the candidate elements, select a candidate element from the expanded exploration nodes according to the report information recorded in the node information of the expanded exploration nodes in the exploration tree, and select a candidate element as a child exploration node of any exploration node.

[0131] Optionally, the node information of the exploration nodes in the exploration tree includes: node effectiveness information, which indicates whether the exploration node is effective. The preset node selection strategy further includes: selecting based on the node effectiveness information when a candidate element is used as an exploration node. Correspondingly, the device may further include:

[0132] The effectiveness evaluation module is configured to evaluate the effectiveness of the factor generation operator corresponding to the exploration node using a preset node effectiveness evaluation strategy, and obtain the effectiveness evaluation information of the exploration node.

[0133] The effectiveness backpropagation module is configured to propagate the node effectiveness information of the exploration node from the exploration node toward the root node of the exploration tree.

[0134] Accordingly, the exploration tree expansion module 902 can be configured to select a valid candidate element for a node when selecting a child exploration node for any exploration node of the exploration tree, if a candidate element is selected from the expanded exploration nodes of the exploration tree.

[0135] Optionally, the node information of the exploration nodes in the exploration tree includes: exploration completeness information, which indicates whether the exploration node has completed exploration. Correspondingly, the preset node selection strategy further includes: selecting based on the exploration completeness information when a candidate element is used as an exploration node. Correspondingly, the device further includes:

[0136] The evaluation judgment module is configured to determine whether the factor trees corresponding to all leaf nodes extended from the exploration node have been evaluated.

[0137] The exploration information update module is configured to determine that the exploration node has completed exploration if the evaluation judgment module determines that it is, and update the exploration completeness information in the node information of the exploration node accordingly.

[0138] The exploration information backpropagation module is configured to propagate the exploration completeness information of the exploration node from the exploration node toward the root node of the exploration tree;

[0139] Accordingly, the exploration tree expansion module 902 can be configured to select a candidate element that has not yet been fully explored when selecting a child exploration node for any exploration node of the exploration tree, based on the exploration completeness information recorded in the node record area of ​​the expanded exploration node.

[0140] Optionally, the exploration tree expansion module 902 can be configured to calculate the score corresponding to the reward information when a candidate element is used as an exploration node based on the upper limit confidence interval algorithm, and select the exploration tree node according to the score.

[0141] Optionally, the exploration tree expansion module 902 is configured to expand the exploration tree in a depth-first manner.

[0142] Optionally, the operator information of the factor generation operator includes constraint information of the input values. Correspondingly, the preset node selection strategy further includes: when selecting a sub-exploration node, selecting candidate elements as sub-exploration nodes based on the constraint information that the output value of the sub-exploration node should satisfy the input value of its parent node.

[0143] Optionally, the operator information of the factor generation operator further includes: operator identification information, execution function information, and input value quantity requirement information. The operator identification information is used to identify the factor generation operator, the execution function information is the function information called when the factor generation operator is evaluated, and the input value quantity requirement information is the number of input values ​​required by the factor generation operator.

[0144] Optionally, the exploration tree is an exploration tree expanded according to the Monte Carlo tree algorithm. Accordingly, the exploration tree expansion module 902 may include:

[0145] The node selection submodule is configured to select candidate elements in a depth-first manner based on the constraint information that the output value of the sub-exploration node should satisfy the input value of the parent node of the sub-exploration node, and to calculate the score corresponding to the reward information when the candidate element is used as an exploration node based on the upper confidence interval algorithm, and select exploration nodes based on the score to expand the exploration tree;

[0146] The information generation submodule is configured to generate node information for any non-leaf node in the factor tree, based on the operator identifier information, execution function information, input value quantity requirement information, and constraint information of the factor generation operator of the non-leaf node. The node information includes: node identifier information, execution function information, numerical quantity requirement information, parent node information, child node information, and whether it has been evaluated.

[0147] The growth determination submodule is configured to determine that the factor tree has completed growth when the number of child nodes of all non-leaf nodes in the factor tree meets the input value requirement information of the non-leaf nodes.

[0148] As can be seen, the aforementioned factor mining device expands the exploration tree through a heuristic exploration algorithm, continuously selecting candidate elements and expanding exploration nodes. During the expansion of the exploration tree, factor trees grow from the exploration nodes. The exploration process of the factor trees is guided by the exploration tree, and the exploration tree and factor trees grow simultaneously. This avoids the population aging problem caused by insufficient diversity with iteration in genetic algorithms. Applying the exploration tree to factor mining based on factor generation operator expressions improves search efficiency and diversity, resulting in better mining effects.

[0149] The above is a schematic diagram of a factor mining device according to this embodiment. It should be noted that the technical solution of this factor mining device and the technical solution of the factor mining method described above belong to the same concept. For details not described in detail in the technical solution of the factor mining device, please refer to the description of the technical solution of the factor mining method described above.

[0150] See Figure 10 , Figure 10 A flowchart illustrating a method for constructing an investment model according to one embodiment of this specification is shown. Figure 10 As shown, the method for constructing the investment model specifically includes the following steps.

[0151] Step 1002: Predefine candidate elements to obtain a set of candidate elements, which include: alpha factor generation operator and quantity-price data.

[0152] Step 1004: Use the factor mining method as described in any of the above embodiments and the set of candidate elements to perform factor mining to obtain a factor tree that satisfies the preset return conditions.

[0153] Step 1006: Construct an investment model based on the alpha factor tree that meets the preset return conditions.

[0154] It should be noted that the embodiments in this specification do not limit the specific means of constructing the investment model. For example, in one or more embodiments, the investment model is a machine learning model based on a neural network, which can use the factor tree that meets the preset return conditions as samples and the corresponding investment returns as labels to train the investment model and obtain the trained investment model.

[0155] The above is an illustrative scheme of a method for constructing an investment model according to this embodiment. It should be noted that the technical solution of this method for constructing an investment model belongs to the same concept as the technical solution of the above-mentioned method for mining factors. For details not described in detail in the technical solution of the method for constructing an investment model, please refer to the description of the technical solution of the above-mentioned method for mining factors.

[0156] See Figure 11 , Figure 11 A structural diagram of an apparatus for constructing an investment model according to one embodiment of this specification is shown. Figure 11 As shown, the apparatus for constructing the investment model specifically includes:

[0157] The element predefinition module 1102 is configured to predefine candidate elements and obtain a set of candidate elements, which include: alpha factor generation operator and quantity-price data.

[0158] The mining module 1104 is configured to perform factor mining using the factor mining method as described in any of the above embodiments and the set of candidate elements to obtain a factor tree that satisfies preset return conditions.

[0159] The model building module 1106 is configured to build an investment model based on the factor tree that meets the preset return conditions.

[0160] The above is an illustrative scheme of an apparatus for constructing an investment model according to this embodiment. It should be noted that the technical solution of this apparatus for constructing an investment model and the technical solution of the method for constructing an investment model described above belong to the same concept. For details not described in detail in the technical solution of the apparatus for constructing an investment model, please refer to the description of the technical solution of the method for constructing an investment model described above.

[0161] See Figure 12 , Figure 12 A flowchart illustrating an embodiment of the interaction method provided in this specification is shown. Figure 12 As shown, the interaction method specifically includes:

[0162] Step 1202: Provide a candidate element selection interface, which includes a scene selection component and a candidate element display component.

[0163] The candidate element selection interface is a graphical user interface used to help users select candidate elements according to the needs of actual application scenarios. The candidate element selection interface may contain various components, including at least a scenario selection component and a candidate element display component. The interface display format and input / output methods of the candidate element selection interface can be set according to the actual scenario requirements. For example, users can use input devices such as a mouse, touchscreen, or voice to manipulate icons or menus on the screen to input selection commands, call information, start program instructions, or perform other tasks.

[0164] The application environment of the interaction method provided in the embodiments of this specification is not limited. For example, when the interaction method is applied to a client, the candidate element selection interface mentioned in the interaction method can be represented as the client's interface. As another example, when the interaction method is applied to a server, the interface mentioned in the interaction method can be represented as the server's interface. Yet another example, when the interaction method is applied to a webpage, the interface mentioned in the interaction method can be represented as a webpage. For example, such as... Figure 13 The diagram shows an interactive interface, where the candidate element selection interface is displayed on the server-side monitor used for factor mining.

[0165] The scene selection component is used to receive a scene selected by the user. The display effect of the scene selection component is not limited and can be set according to the needs of the application scenario. For example, the scene selection component may include any selection component such as a radio button, a radio list, or a draggable slider. When the scene selection component is used by the user, it can be displayed as a selected or unselected state for any preset scene. Figure 13 The interactive interface diagram shown depicts a drop-down menu corresponding to the "Scene Selection" text box. This drop-down menu can be understood as the scene selection component. Multiple preset scenes can be displayed in this drop-down menu, each with a corresponding radio button. Users can select the corresponding preset scene by clicking any radio button.

[0166] The candidate element display component is used to display a list of candidate elements corresponding to the scenario selected by the user. The display effect of the candidate element display component is not limited and can be set according to the needs of the application scenario. For example, the candidate element display component may include any display component such as a checklist or a text box. When any preset scenario is selected, the candidate element display component displays a list of factor generation operators corresponding to that preset scenario. Figure 13The interactive interface diagram shown displays a candidate element list to the right of "Scene Selection." The display area below this candidate element list can be understood as a candidate element display component. Users can directly select all factor generation operators in the candidate element list for factor mining, or further select one or more factor generation operators as needed by clicking checkboxes. Additionally, depending on the actual application scenario, a "Add Candidate Element" component can be provided. When the user triggers the "Add Candidate Element" component on the interface, an input area for adding a new candidate element will be displayed, allowing the user to input custom factor generation operators as needed.

[0167] Step 1204: In response to the user selecting any preset scene using the scene selection component in the candidate element selection interface, the list of factor generation operators corresponding to any preset scene is displayed through the candidate element display component.

[0168] For example, multiple preset scenarios can be set according to actual application needs, such as investment prediction scenarios, content push scenarios, etc. Factor generation operators can be pre-set for these preset scenarios according to their specific requirements. The lists of factor generation operators for different preset scenarios can be the same or different, but at least two preset scenarios must have different lists of factor generation operators.

[0169] Step 1206: Obtain the factor generation operator selected by the user from the list of factor generation operators and obtain the initial features input by the user as a set of candidate elements.

[0170] The implementation method for user input of initial features is not limited. For example, the interactive interface can provide an operation component for adding initial features, through which the user can input initial features from the outside; as another example, the interactive interface can provide an operation component for selecting initial features, through which the user can select some initial features provided internally by the system; yet another example is that, in addition to providing an operation component for selecting initial features, the user can also input initial features from the outside through the operation component for adding initial features provided internally by the interactive interface.

[0171] Step 1208: Using the factor mining method as described in any embodiment of this specification and the set of candidate elements, perform factor mining to mine a factor tree that satisfies the preset return conditions.

[0172] As can be seen, this interactive method helps users set the corresponding set of candidate elements according to the needs of the scenario, helps users to more conveniently apply the methods provided in the embodiments of this specification to mine factors, and further improves the mining efficiency.

[0173] See Figure 14 , Figure 14 A structural diagram of an interactive device provided in one embodiment of this specification is shown. Figure 14 As shown, the interactive device specifically includes:

[0174] The interface providing module 1402 is configured to provide a candidate element selection interface, which includes a scene selection component and a candidate element display component.

[0175] The scene selection module 1404 is configured to, in response to a user selecting any preset scene using the scene selection component in the candidate element selection interface, display a list of factor generation operators corresponding to any preset scene through the candidate element display component.

[0176] The candidate element acquisition module 1406 is configured to acquire the factor generation operator selected by the user from the factor generation operator list and acquire the initial features input by the user as a set of candidate elements.

[0177] The mining trigger module 1408 is configured to perform factor mining using the factor mining method as described in any embodiment of this specification and the set of candidate elements, and to mine a factor tree that satisfies preset return conditions.

[0178] The above is an illustrative scheme of an interactive device according to this embodiment. It should be noted that the technical solution of this interactive device and the technical solution of the above-described interactive method belong to the same concept. For details not described in detail in the technical solution of the interactive device, please refer to the description of the technical solution of the above-described interactive method.

[0179] Figure 15 A structural block diagram of a computing device 1500 according to one embodiment of this specification is shown. The components of the computing device 1500 include, but are not limited to, a memory 1510 and a processor 1520. The processor 1520 is connected to the memory 1510 via a bus 1530, and a database 1550 is used to store data.

[0180] The computing device 1500 also includes an access device 1540, which enables the computing device 1500 to communicate via one or more networks 1560. Examples of these networks include Public Switched Telephone Network (PSTN), Local Area Network (LAN), Wide Area Network (WAN), Personal Area Network (PAN), or combinations of communication networks such as the Internet. The access device 1540 may include one or more of any type of wired or wireless network interface (e.g., a network interface card (NIC)), such as an IEEE 802.11 Wireless Local Area Network (WLAN) wireless interface, a Wi-MAX (Worldwide Interoperability for Microwave Access) interface, an Ethernet interface, a Universal Serial Bus (USB) interface, a cellular network interface, a Bluetooth interface, or a Near Field Communication (NFC) interface.

[0181] In one embodiment of this specification, the above-described components of the computing device 1500 and Figure 15 Other components, not shown, can also be connected to each other, for example, via a bus. It should be understood that... Figure 15 The block diagram of the computing device shown is for illustrative purposes only and is not intended to limit the scope of this specification. Those skilled in the art can add or replace other components as needed.

[0182] The computing device 1500 can be any type of stationary or mobile computing device, including mobile computers or mobile computing devices (e.g., tablet computers, personal digital assistants, laptop computers, notebook computers, netbooks, etc.), mobile phones (e.g., smartphones), wearable computing devices (e.g., smartwatches, smart glasses, etc.) or other types of mobile devices, or stationary computing devices such as desktop computers or personal computers (PCs). The computing device 1500 can also be a mobile or stationary server.

[0183] The processor 1520 is configured to execute the following computer-executable instructions, which, when executed by the processor, implement the above-mentioned method for mining factors, or implement the above-mentioned method for constructing an investment model, or implement the steps of the above-mentioned interactive method.

[0184] The above is an illustrative scheme of a computing device according to this embodiment. It should be noted that the technical solution of this computing device belongs to the same concept as the technical solutions of the above-mentioned method for mining factors, method for constructing investment models, or interactive method. For details not described in detail in the technical solution of the computing device, please refer to the description of the technical solutions of the above-mentioned method for mining factors, method for constructing investment models, or interactive method.

[0185] An embodiment of this specification also provides a computer-readable storage medium storing computer-executable instructions that, when executed by a processor, implement the above-described method for mining factors, or the above-described method for constructing an investment model, or the steps of the above-described interactive method.

[0186] The above is an illustrative scheme of a computer-readable storage medium according to this embodiment. It should be noted that the technical solution of this storage medium belongs to the same concept as the above-described method for mining factors, or the above-described method for constructing investment models, or the above-described interaction method. For details not described in detail in the technical solution of the storage medium, please refer to the description of the technical solution of the above-described method for mining factors, or the above-described method for constructing investment models, or the above-described interaction method.

[0187] An embodiment of this specification also provides a computer program, wherein when the computer program is executed in a computer, it causes the computer to perform the above-described method for mining factors, or the above-described method for constructing an investment model, or the steps of the above-described interactive method.

[0188] The above is an illustrative example of a computer program in this embodiment. It should be noted that the technical solution of this computer program belongs to the same concept as the above-described method for mining factors, or the above-described method for constructing investment models, or the above-described interactive method. Details not described in detail in the computer program's technical solution can be found in the descriptions of the technical solutions for the above-described method for mining factors, or the above-described method for constructing investment models.

[0189] The foregoing has described specific embodiments of this specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims may be performed in a different order than that shown in the embodiments and may still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require the specific or sequential order shown to achieve the desired result. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

[0190] The computer instructions include computer program code, which may be in the form of source code, object code, executable file, or certain intermediate forms. The computer-readable medium may include any entity or device capable of carrying the computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in the computer-readable medium may be appropriately added to or subtracted according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media may not include electrical carrier signals and telecommunication signals.

[0191] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that the embodiments in this specification are not limited to the described order of actions, because according to the embodiments in this specification, some steps can be performed in other orders or simultaneously. Furthermore, those skilled in the art should also understand that the embodiments described in this specification are all preferred embodiments, and the actions and modules involved are not necessarily essential to the embodiments in this specification.

[0192] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0193] The preferred embodiments disclosed above are merely illustrative of this specification. The optional embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. Clearly, many modifications and variations can be made based on the embodiments described herein. These embodiments are selected and specifically described in this specification to better explain the principles and practical applications of the embodiments, thereby enabling those skilled in the art to better understand and utilize this specification. This specification is limited only by the claims and their full scope and equivalents.

Claims

1. An interaction method, comprising: A candidate element selection interface is provided, which includes a scene selection component and a candidate element display component; In response to a user selecting any preset scene using the scene selection component in the candidate element selection interface, the candidate element display component displays a list of factor generation operators corresponding to any preset scene. Obtain the factor generation operator selected by the user from the list of factor generation operators and obtain the initial features input by the user as a set of candidate elements; Factor mining is performed using a factor mining method and the set of candidate elements to mine a factor tree that satisfies preset return conditions. The factor mining method includes: Candidate elements are selected as exploration nodes based on a preset node selection strategy to expand the exploration tree. The preset node selection strategy includes: selecting based on the reward information when candidate elements are used as exploration nodes. The candidate elements include: factor generation operators and initial features. A factor tree is obtained from the exploration tree, wherein the factor tree is a complete subtree extended from the child exploration node under the root node of the exploration tree, and the factor generation operator is a non-leaf node and the initial feature is a leaf node in the complete subtree; The factor generation operators of the non-leaf nodes in the factor tree are evaluated using the initial features of the leaf nodes in the factor tree to determine the factor values ​​corresponding to the factor tree. Using the factor values ​​of the factor tree and a preset reward information evaluation strategy, the reward information of the factor tree is evaluated, and the reward information of the factor tree is propagated from the leaf node of the factor tree to the root node of the exploration tree to update the reward information of the exploration nodes on the propagation path. Based on the return information of each factor tree, determine the factor trees that meet the preset return conditions.

2. The method according to claim 1, wherein the node information of the exploration nodes in the exploration tree includes: The reward information from the exploration node; The selection based on the reward information when candidate elements are explored includes: When selecting a child exploration node for any exploration node, if there are remaining candidate elements that have not yet been selected among the candidate elements, then a candidate element is selected from the remaining candidate elements as a child exploration node of any exploration node. If there are no remaining unselected candidate elements among the candidate elements, then candidate elements are selected from the expanded exploration nodes according to the report information recorded in the node information of the expanded exploration nodes of the exploration tree, and used as child exploration nodes of any of the exploration nodes.

3. The method according to claim 1, wherein the node information of the exploration nodes in the exploration tree includes: Node validity information, which indicates whether the exploration node is valid; The preset node selection strategy also includes: selecting based on the node effectiveness information when candidate elements are explored nodes; The method further includes: The effectiveness of the factor generation operator corresponding to the exploration node is evaluated using a preset node effectiveness evaluation strategy to obtain the effectiveness evaluation information of the exploration node. The node validity information of the exploration node is propagated from the exploration node toward the root node of the exploration tree; The selection based on the node effectiveness information when candidate elements are explored includes: When selecting a child exploration node for any exploration node in the exploration tree, if a candidate element is selected from the expanded exploration nodes of the exploration tree, then a valid candidate element is selected based on the node validity information recorded in the node record area of ​​the expanded exploration node.

4. The method according to claim 1, wherein the node information of the exploration nodes in the exploration tree includes: Explore completeness information, which indicates whether the exploration node has completed the exploration; The preset node selection strategy also includes: selecting based on the exploration completeness information when candidate elements are used as exploration nodes; The method further includes: Determine whether the factor trees corresponding to all leaf nodes extended from the exploration node have been evaluated; If so, then determine that the exploration node has completed the exploration, and update the exploration completeness information in the node information of the exploration node accordingly; The exploration completeness information of the exploration node is propagated from the exploration node toward the root node of the exploration tree; The selection based on the exploration completeness information when candidate elements are used as exploration nodes includes: When selecting a child exploration node for any exploration node in the exploration tree, if a candidate element is selected from the exploration nodes that have been expanded in the exploration tree, then the candidate element that has not yet been fully explored is selected based on the exploration completeness information recorded in the node record area of ​​the expanded exploration node.

5. The method according to claim 1, wherein selecting based on the reward information when a candidate element is used as an exploration node includes: The upper limit confidence interval algorithm is used to calculate the score corresponding to the reward information when a candidate element is used as an exploration node, and the exploration tree node is selected based on the score.

6. The method according to claim 1, wherein the expanded exploration tree comprises: The exploration tree is expanded in a depth-first manner.

7. The method according to any one of claims 1-6, wherein the operator information of the factor generation operator includes constraint information of the input values; The preset node selection strategy also includes: When selecting a child exploration node, candidate elements are selected based on the constraint information that the output value of the child exploration node should satisfy the input value of the parent node of the child exploration node.

8. The method according to claim 7, wherein the operator information of the factor generation operator further includes: The operator identification information, execution function information, and input value quantity requirement information are as follows: the operator identification information is used to identify the factor generation operator; the execution function information is the function information called when the factor generation operator is evaluated; and the input value quantity requirement information is the number of input values ​​required by the factor generation operator.

9. The method according to claim 8, wherein the exploration tree is an exploration tree expanded according to the Monte Carlo tree algorithm; The step of selecting candidate elements as exploration nodes based on a preset node selection strategy and expanding the exploration tree includes: Following a depth-first approach, candidate elements are selected based on the constraint that the output value of a child exploration node should satisfy the input value of its parent node. Additionally, a score is calculated based on the upper limit confidence interval algorithm to determine the reward information corresponding to the candidate element as an exploration node. Exploration nodes are then selected based on the scores to expand the exploration tree. For any non-leaf node in the factor tree, based on the operator identifier information, execution function information, input value quantity requirement information, and constraint information of the factor generation operator of the non-leaf node, node information of the non-leaf node is generated. The node information includes: node identifier information, execution function information, numerical quantity requirement information, parent node information, child node information, and whether it has been evaluated. When the number of child nodes of all non-leaf nodes in the factor tree meets the requirement for the number of input values ​​of the non-leaf nodes, the factor tree is determined to have completed growth.

10. An interaction method, comprising: A candidate element selection interface is provided, which includes a scene selection component and a candidate element display component; In response to a user selecting any preset scene using the scene selection component in the candidate element selection interface, the candidate element display component displays a list of factor generation operators corresponding to any preset scene. Obtain the factor generation operator selected by the user from the list of factor generation operators and obtain the initial features input by the user as a set of candidate elements; Factor mining is performed using a factor mining method and the set of candidate elements to mine a factor tree that satisfies preset return conditions. The factor mining method includes: Candidate elements are selected as exploration nodes based on a preset node selection strategy to expand the exploration tree. The preset node selection strategy includes: selecting based on the reward information when candidate elements are used as exploration nodes. The candidate elements include: alpha factor generation operator and quantity-price data. An alpha factor tree is obtained from the exploration tree, wherein the alpha factor tree is a complete subtree extended from the child exploration node under the root node of the exploration tree, and the alpha factor generating operator is used as a non-leaf node and the quantity and price data is used as a leaf node in the complete subtree; The alpha factor generation operator of the non-leaf nodes in the alpha factor tree is evaluated using the quantity and price data of the leaf nodes in the alpha factor tree to determine the factor value corresponding to the alpha factor tree. Using the factor values ​​of the alpha factor tree and a preset reward information evaluation strategy, the reward information of the alpha factor tree is evaluated, and the reward information of the alpha factor tree is propagated from the leaf node of the alpha factor tree to the root node of the exploration tree to update the reward information of the exploration nodes on the propagation path. Based on the return information of each alpha factor tree, determine the alpha factor trees that meet the preset return conditions.

11. A method for constructing an investment model, comprising: Candidate elements are predefined to obtain a set of candidate elements, which include: alpha factor generation operator and quantity-price data; Factor mining is performed using the method described in claim 10 and the set of candidate elements to obtain an alpha factor tree that satisfies the preset return conditions; An investment model is constructed based on the alpha factor tree that meets the preset return conditions.

12. A computing device, comprising: Memory and processor; The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, they implement the steps of the interactive method according to any one of claims 1 to 10, or implement the steps of the method for constructing an investment model according to claim 11.

13. A computer-readable storage medium storing computer-executable instructions that, when executed by a processor, implement the steps of the interactive method of any one of claims 1 to 10, or the steps of the method for constructing an investment model of claim 11.

Images