Data fusion system and data fusion method

Abstract

The invention provides a data fusion system and a data fusion method. The data fusion system comprises a storage device and a processor. The storage device is used for storing a feature acquisition module, a model construction module, a model disassembly module and a data fusion module. The processor is electrically connected with the storage device. The processor is used for executing the feature acquisition module, the model construction module, the model disassembly module and the data fusion module, and is used for receiving a plurality of pieces of form data. The feature capturing module captures a plurality of feature data from a plurality of form data. The model construction module performs preprocessing operation on the plurality of feature data and constructs a prediction model. The model splitting module splits the prediction model into a plurality of triple data. And the data fusion module fuses the multiple pieces of triple data into the knowledge graph data. According to the data fusion system and the data fusion method, the redundancy and the complexity after data amplification can be effectively reduced.

Description

Technical Field

[0001] The invention relates to a data processing technology, in particular to a data fusion system and a data fusion method. Background Art

[0002] Current Enterprise Resource Planning (ERP) systems can use pre-built forecasting models to automatically forecast business data, and the forecasting models can be built based on relevant historical business data and / or relevant historical information. However, since the amount of relevant historical business data and / or relevant historical information is huge and may contain a large amount of redundant data, the complexity of the forecasting model may increase, resulting in poor system operation efficiency, poor forecasting accuracy, and / or excessive computing resources consumed by the system. Summary of the invention

[0003] The present invention is directed to a data fusion system and a data fusion method, which can effectively reduce the redundancy and complexity of data after amplification.

[0004] According to an embodiment of the present invention, the data fusion system of the present invention includes a storage device and a processor. The storage device is used to store a feature extraction module, a model construction module, a model disassembly module and a data fusion module. The processor is electrically connected to the storage device, and is used to execute the feature extraction module, the model construction module, the model disassembly module and the data fusion module, and is used to receive multiple form data. The feature extraction module extracts multiple feature data from multiple form data. The model construction module performs preprocessing operations on the multiple feature data and constructs a prediction model. The model disassembly module disassembles the prediction model into multiple triple data. The data fusion module fuses multiple triple data into the knowledge graph data.

[0005] According to an embodiment of the present invention, the data fusion method of the present invention includes the following steps: executing a feature extraction module through a processor to extract multiple feature data from multiple form data; executing a model construction module through a processor to perform preprocessing operations on multiple feature data and build a prediction model; executing a model disassembly module through a processor to disassemble the prediction model into multiple triple data; and executing a data fusion module through a processor to fuse multiple triple data into knowledge graph data.

[0006] Based on the above, the data fusion system and data fusion method of the present invention can convert form data into feature data to construct a prediction model, and can fuse the data of the prediction model into the knowledge graph data executed by the system to effectively reduce the redundancy and complexity after data amplification, and thereby enable the system to realize effective business prediction functions based on the knowledge graph data.

[0007] In order to make the above features and advantages of the present invention more clearly understood, embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] Figure 1 is a schematic diagram of a data fusion system according to an embodiment of the present invention;

[0009] Figure 2 is a flow chart of a data fusion method according to an embodiment of the present invention;

[0010] Figure 3 is a schematic diagram of data conversion according to an embodiment of the present invention;

[0011] Figure 4 is a schematic diagram of a prediction model according to an embodiment of the present invention;

[0012] Figure 5 It is a schematic diagram of a knowledge graph according to an embodiment of the present invention.

[0013] Description of Reference Numerals

[0014] 100: Data fusion system;

[0015] 110: processor;

[0016] 120: storage device;

[0017] 121: feature extraction module;

[0018] 122: Model building module;

[0019] 123: Model disassembly module;

[0020] 124: data fusion module;

[0021] 210: Electronic devices;

[0022] 220: Database;

[0023] 301: Historical purchase form;

[0024] 302: Purchase form;

[0025] 303: Supplier basic information form;

[0026] 304: Summary table data;

[0027] 305: Dataset;

[0028] 306: sample data;

[0029] 307: sample label pair set;

[0030] 308: Prediction model;

[0031] 309: triple data;

[0032] 310: Knowledge graph data;

[0033] 401-421, 511, 521, 522, 531-538, 541, 542: nodes;

[0034] S210~S240: steps. DETAILED DESCRIPTION

[0035] Reference will now be made in detail to exemplary embodiments of the present invention, examples of which are illustrated in the accompanying drawings. Whenever possible, the same reference numerals are used in the drawings and the description to refer to the same or like parts.

[0036] Figure 1 FIG. 1 is a schematic diagram of a data fusion system according to an embodiment of the present invention. Figure 1 , the data fusion system 100 includes a processor 110 and a storage device 120. The processor 110 is electrically connected to the storage device 120. The storage device 120 can store a feature extraction module 121, a model construction module 122, a model disassembly module 123, and a data fusion module 124. In this embodiment, the data fusion system 100 can be implemented by a cloud server (Server) to provide a software as a service (Software as a service) function, or by a computer device, and the data fusion system 100 can communicate with an electronic device 210 and a database 220. In this embodiment, the processor 110 can obtain a plurality of form data and / or related business data from the database 220, and perform data processing on the plurality of form data and / or related business data to fuse and expand them into the knowledge graph data.

[0037] The electronic device 210 may be, for example, a user device. The processor 110 may obtain input data (i.e., query data) from the electronic device 210, and may query the amplified knowledge graph data according to the input data to feed back corresponding prediction results to the electronic device 210. In this embodiment, the data fusion system 100 may be used to implement abnormal supplier replacement prediction in the manufacturing industry, but the present invention is not limited thereto.

[0038] In the present embodiment, the processor 110 may be a system on a chip (SOC), or may include, for example, a central processing unit (CPU) or other programmable general-purpose or special-purpose microprocessor, digital signal processor (DSP), programmable controller, application specific integrated circuits (ASIC), programmable logic device (PLD), other similar processing devices or a combination of these devices. In the present embodiment, the storage device 120 may be, for example, a dynamic random access memory (DRAM), a flash memory or a non-volatile random access memory (NVRAM). In the present embodiment, the storage device 120 may be used to store the relevant data and related information described in the various embodiments of the present invention. The storage device 120 can be used to store the algorithms and programs of the feature extraction module 121, the model construction module 122, the model disassembly module 123, and the data fusion module 124 for access and / or execution by the processor 110. The storage device 120 can also store or temporarily store the relevant data and knowledge graph required or generated during the operation of the data fusion system 100.

[0039] In this embodiment, the electronic device 210 may be, for example, a user computer or an input device (e.g., a keyboard, a mouse, or a touch interface, etc., with a physical or virtual input port). The database 220 may be, for example, used to store relevant historical business data, historical business forms, and / or relevant historical information related to an enterprise resource planning (ERP) system. In addition, it should be noted that the knowledge graph described in the embodiments of the present invention is a graphical database for organizing and representing knowledge. It contains relevant business information and the relationships between them. The knowledge graph is a data structure used to describe the associations between real business data so that the processor (computer) can perform automated logical reasoning based on this data structure.

[0040] Figure 2 is a flow chart of a data fusion method according to an embodiment of the present invention. Figure 3 FIG. 1 is a schematic diagram of data conversion according to an embodiment of the present invention. Figures 1 to 3, the data fusion system 100 may execute the following steps S210 to S240. In step S210, the processor 110 may execute the feature extraction module 121 to extract multiple feature data from multiple form data. In this embodiment, the processor 110 may, for example, obtain multiple form data from a database, wherein the multiple form data may, for example, include: Figure 3 At least one of the historical purchase form 301, the purchase form 302, and the supplier basic information form 303 shown. In this embodiment, the feature extraction module 121 can perform table association processing on multiple form data to generate summary table data 304, and perform ETL (Extract, Transform, Load) engineering calculation on the summary table data to generate multiple feature data. The processor 110 can store the multiple feature data as a data set 305 for subsequent modeling. Table association processing means that the feature extraction module 121 can associate the purchase form with the purchase form according to the table fields purchase order category, purchase order number and serial number, and then use the table field manufacturer code to associate it with the supplier basic information form.

[0041] In step S220, the processor 110 may execute the model building module 122 to perform preprocessing operations on multiple feature data and build a prediction model. In this embodiment, the preprocessing operation includes data removal and data filling on multiple feature data, and selecting relevant features according to analysis of variance (ANOVA) to generate multiple sample data 306, and establishing a sample label pair set 307 based on the multiple sample data 306. For example, the multiple sample data 306 can be used to describe the number of days of delay respectively. The model building module 122 can divide the multiple sample data 306 according to whether the number of days of delay is within 7 days, and set the corresponding labels to 0 or 1 respectively. In this embodiment, the model building module 122 can build a prediction model 308 based on the sample label pair set 307 and a decision tree algorithm. In addition, data removal refers to removing abnormal data. For example, if all features of the data are null values, it means that the data is abnormal data and needs to be removed. Data filling refers to the need to fill missing values ​​in some features of certain data. In this regard, the specific method of data filling may be, for example, mean filling, mode filling or interpolation, and the present invention is not limited thereto.

[0042] In step S230, the processor 110 may execute the model disassembly module 123 to disassemble the prediction model 308 into a plurality of triples (Triples Data) 309. In this embodiment, the prediction model 308 may be a tree structure, and each node data of the tree structure includes information entropy, sample quantity information, category quantity information and category information. The triples data 309 may be composed of factor data, condition data and conclusion data. Specifically, the factor refers to the set of all features on the path in the decision tree, the condition is the set of all judgment conditions on the path, and the conclusion is the predicted category of this path.

[0043] In step S240, the processor 110 may execute the data fusion module 124 to fuse the plurality of triple data 309 into the knowledge graph data 310. In this embodiment, the data fusion module 124 may fuse the plurality of triple data 309 into the knowledge graph data 310 according to the set relationship between the factor data and the condition data of the plurality of triple data 309, and according to the data size relationship between the conclusion data of the plurality of triple data 309. It should be noted that the data fusion module 124 may merge at least a portion of the plurality of triple data and fuse them into the knowledge graph data 310.

[0044] In this embodiment, the processor 110 may receive input data from the electronic device 210, and the processor 110 may search the knowledge graph data 310 according to the input data to generate a prediction result. Taking the data fusion system 100 for implementing abnormal supplier replacement prediction in the manufacturing industry as an example, the prediction result may include, for example, the number of days of delay or the number of days of order delivery.

[0045] Figure 4 FIG. 1 is a schematic diagram of a prediction model according to an embodiment of the present invention. Figure 1 as well as Figure 4 The data structure of the prediction model described in the above embodiment can be as follows: Figure 4 shown. Figure 4The data structure of the decision tree of the delivery delay prediction model for a certain year may be generated according to the purchase data, and includes a plurality of nodes 401-421 corresponding to a plurality of decision features. In this embodiment, node 401 may be a first-layer node, and may include conditional data indicators "special<=0.383" (i.e., the dividing point of the feature special in the decision tree), information entropy "entropy=0.755" (used to measure the purity of the sample set), sample quantity information "samples=3415" (referring to the number of samples in this set, i.e., the total number of samples after the training set is divided according to a certain feature in the decision tree, for example, the number of all samples of the feature special<=0.383 in the training set), category quantity information "value=[2674,741]" (referring to the number of positive and negative class samples in this set, i.e., the number of delayed and non-delayed samples), and category information "class=0" (class is a category label, and since the number of delayed samples 2674 in Value=[2674,741] is greater than the number of non-delayed samples 741, it is classified as delayed, and its category label is 0). In this regard, if the condition is met (yes), the path points to node 402. If the condition is not met (no), the path points to node 403. Nodes 402 and 403 may be second-layer nodes.

[0046] Node 402 may include the conditional data indicator "o_order<=555.23" (o_order refers to the number of orders in the last month, which is the judgment condition in the decision tree and is used to determine whether the number of orders in the last month is less than or equal to 555.23), information entropy "entropy=0.735", sample quantity information "samples=3198", category quantity information "value=[2538,660]" and category information "class=0". In this regard, if the condition is met (yes), the path points to node 404. If the condition is not met (no), the path points to node 405. Node 403 may include the conditional data indicator "td_t_special<=0.098" (this is the judgment condition in the decision tree and is used to determine the number of feature special less than or equal to 0.098), information entropy "entropy=0.983", sample quantity information "samples=217", category quantity information "value=[136,81]" and category information "class=0". In this regard, if the condition is met (yes), the path points to node 406. If the condition is not met (no), the path points to node 407. Nodes 404 to 407 may be third-layer nodes.

[0047] Node 404 may include the conditional data indicator "td_s_success <= 2369.894" (td_s_success refers to the number of successes in the last 6 months, which is a judgment condition in the decision tree and can be used to determine whether the number of successes in the last 6 months is less than or equal to 2369.894), information entropy "entropy = 0.748", sample quantity information "samples = 3054", category quantity information "value = [2402, 652]" and category information "class = 0". In this regard, if the condition is met (yes), the path points to node 408. If the condition is not met (no), the path points to node 409. Node 405 may include the conditional data indicator "td_o_price_std <= 6603.214" (td_o_price_std refers to the standard deviation of the price in the last month, which is the judgment condition in the decision tree and can be used to judge whether the standard deviation of the price in the last month is less than or equal to 6603.21), information entropy "entropy = 0.31", sample quantity information "samples = 114", category quantity information "value = [136, 8]" and category information "class = 0". In this regard, if the condition is met (yes), the path points to node 410. If the condition is not met (no), the path points to node 411. Node 406 may include the conditional data indicator "td_s_success <= 5.334" (td_s_success refers to the number of successes in the last 6 months, which is a judgment condition in the decision tree and can be used to determine whether the number of successes in the last 6 months is less than or equal to 5.334), information entropy "entropy = 0.973", sample quantity information "samples = 201", category quantity information "value = [120, 81]" and category information "class = 0". In this regard, if the condition is met (yes), the path points to node 412. If the condition is not met (no), the path points to node 413. Node 407 may be an end node and includes information entropy "entropy = 0.0", sample quantity information "samples = 16", category quantity information "value = [16, 0]" and category information "class = 0". Nodes 408 to 413 may be fourth-layer nodes.

[0048] Node 408 may include the conditional data indicator "td_s_success <= 202.942" (td_s_success refers to the number of successes in the last 6 months, which is a judgment condition in the decision tree and can be used to determine whether the number of successes in the last 6 months is less than or equal to 202.94), information entropy "entropy = 0.752", sample quantity information "samples = 3003", category quantity information "value = [2355, 648]" and category information "class = 0". In this regard, if the condition is met (yes), the path points to node 414. If the condition is not met (no), the path points to node 415. Node 409 may include the data indicator of the condition "td_o_price_std<=1241.0" (td_o_price_std refers to the standard deviation of the price in the last month, which is the judgment condition in the decision tree and can be used to determine whether the standard deviation of the price in the last month is less than or equal to 1241.0), information entropy "entropy=0.397", sample quantity information "samples=51", category quantity information "value=[47,4]" and category information "class=0". In this regard, if the condition is met (yes), the path points to node 416. If the condition is not met (no), the path points to node 417. Node 410 may be an end node and may include information entropy "entropy=0.0", sample quantity information "samples=74", category quantity information "value=[74,0]" and category information "class=0". Node 411 may include the conditional data indicator "s_price<=210.546" (s_price refers to the average price of the last 6 months, which is the judgment condition in the decision tree and can be used to judge whether the average price of the last 6 months is less than or equal to 210.546), information entropy "entropy=0.513", sample quantity information "samples=70", category quantity information "value=[62,8]", and category information "class=0". In this regard, if the condition is met (yes), the path points to node 418. If the condition is not met (no), the path points to node 419. Node 412 may include the conditional data indicator "td_o_order<=3.834" (td_o_order refers to the number of orders in the last month, which is a judgment condition in the decision tree and can be used to determine whether the number of orders in the last month is less than or equal to 3.834), information entropy "entropy=0.977", sample quantity information "samples=197", category quantity information "value=[116,81]", and category information "class=0". In this regard, if the condition is met (yes), the path points to node 420. If the condition is not met (no), the path points to node 421.Node 413 may be a terminal node, and may include information entropy “entropy=0.0”, sample quantity information “samples=4”, category quantity information “value=[4,0]”, and category information “class=0”. Nodes 414 to 421 may be fifth-layer nodes.

[0049] Node 414 may be a terminal node and may include information entropy "entropy = 0.72", sample quantity information "samples = 2241", category quantity information "value = [1795, 446]", and category information "class = 0". Node 415 may be a terminal node and may include information entropy "entropy = 0.834", sample quantity information "samples = 762", category quantity information "value = [650, 202]", and category information "class = 0". Node 416 may be a terminal node and may include information entropy "entropy = 0.918", sample quantity information "samples = 3", category quantity information "value = [1, 2]", and category information "class = 1". Node 417 may be a terminal node and may include information entropy "entropy = 0.25", sample quantity information "samples = 48", category quantity information "value = [46, 2]", and category information "class = 0". Node 418 may be a terminal node and may include information entropy “entropy=0.0”, sample quantity information “samples=35”, category quantity information “value=[35,0]”, and category information “class=0”. Node 419 may be a terminal node and may include information entropy “entropy=0.776”, sample quantity information “samples=35”, category quantity information “value=[27,8]”, and category information “class=0”. Node 420 may be a terminal node and may include information entropy “entropy=0.976”, sample quantity information “samples=196”, category quantity information “value=[116,80]”, and category information “class=0”. Node 421 may be a terminal node and may include information entropy “entropy=0.0”, sample quantity information “samples=1”, category quantity information “value=[0,1]”, and category information “class=1”.

[0050] In this embodiment, the model construction module 122 can divide the procurement data by year to construct a corresponding plurality of prediction models, and the model disassembly module 123 can disassemble the plurality of prediction models to generate a plurality of triple data. In this regard, the data fusion module 124 can generate corresponding knowledge graphs according to the triple data, and can merge and fuse the graphs by year. For example, the data fusion system 100 can construct a first knowledge graph using the business data of 2011, and a second knowledge graph using the business data of 2012. The data fusion system 100 can fuse the first knowledge graph and the second knowledge graph using the fusion rule to obtain a third knowledge graph, and then fuse the third knowledge graph with the fourth knowledge graph constructed by the business data of the next year. In this way, the prediction performance of the knowledge graph can be continuously improved as the knowledge graph is continuously fused.

[0051] In this embodiment, the fusion of the knowledge graph is to fuse triples using fusion rules. The fusion rules are mainly composed of the set relationship between the factor data and condition data of multiple triple data, and the data size relationship between the conclusion data of multiple triple data to fuse the multiple triple data into the knowledge graph data. For example, the factor data of two triple data are consistent, and there is only one difference in the conditions corresponding to all factors, where the intersection of different conditions is an empty set, and the absolute value of the difference in the probabilities of the corresponding delay situations is less than the threshold value of 0.25, for example, the probability values ​​are all greater than 0.5, or are less than or equal to 0.5.

[0052] For another example, the two triplets in Table 1 below can be merged into a new triplet, and the two triplets in Table 2 below cannot be merged. In this regard, in Table 1, since the probabilities of the two triplets are both greater than 0.5, they are judged to be in the same category, and their conditions can be merged. However, in Table 2, since the probabilities of the two triplets are one greater than 0.5 and the other less than 0.5, they are judged not to belong to the same category and cannot be classified into one category. In this regard, the new triplet in Table 1 will be merged into the knowledge graph, and the two triplets in Table 2 will expand the knowledge graph.

[0053]

[0054] Table 1

[0055]

[0056] Table 2

[0057] Figure 5 Schematic diagram of a knowledge graph according to an embodiment of the present invention. Figure 1 as well as Figure 5, the knowledge graph data of the present invention may be, for example, Figure 5 In this embodiment, Figure 5 The knowledge graph may include nodes 511, 521-522, 531-538, 541-542. Node 511 is used to represent the supplier selection mechanism. Nodes 521 and 522 correspond to factor data. Node 521 is used to represent "number of orders in the past month / whether special purchase / number of orders in the past month (materials) / number of successful deliveries in the past 6 months (materials)". Node 522 is used to represent "whether special purchase / number of orders in the past month (materials) / number of successful deliveries in the past 6 months (materials)".

[0058] Nodes 531 to 538 correspond to conditional data. Node 531 is used to indicate "whether special purchase <= 0.383 / number of orders in the past month <= 555.23 / number of successful deliveries in the past 6 months (materials)". Node 532 is used to indicate "number of orders in the past month <= 555.23 / number of successful deliveries in the past 6 months (materials)". Node 533 is used to indicate "whether special purchase <= 0.383 / number of orders in the past month > 555.23". Node 534 is used to indicate "whether special purchase > 0.383 / number of special purchases (materials) in the last 3 months> 0.098. Node 535 is used to indicate "whether special purchase > 0.383 / number of special purchases (materials) in the last 3 months<=0.098 / number of successful deliveries in the last 6 months". Node 536 is used to indicate "whether special purchase > 0.383 / number of special purchases (materials) in the last 3 months<=0.098 / number of successful deliveries in the last 6 months". Node 537 is used to indicate "whether special purchase <=0.383 / number of orders in the last month<=555.23 / number of successful deliveries in the last 6 months (materials)". Node 537 is used to indicate "whether special purchase > 0.383 / number of special purchases in the last 3 months<=0.098 / number of successful deliveries in the last 6 months (materials)".

[0059] Nodes 541 to 542 correspond to conclusion data. Node 541 is used to indicate "if the delivery delay is less than 7 days, this supplier can be selected." Node 542 is used to indicate "if the delivery delay is more than 7 days, this supplier cannot be selected."

[0060] In this embodiment, the data fusion system 100 can obtain input data (original data) input by the user and search for Figure 5The knowledge graph of the data is used to predict delivery delays. Specifically, the user can input the input data as shown in Table 3 below, where the input data of Table 3 includes multiple key-value pairs, where the keys refer to features and the values ​​are specific values ​​under each feature. In Table 3, "urgent" refers to whether urgent material procurement is required. "o_success" refers to the number of successful deliveries in the past month. "o_order" refers to the number of orders in the past month. "o_delivery" refers to the average delivery time in the past month. "t_time" refers to the number of on-time deliveries in the past three months. "t_order" refers to the number of orders in the past three months. "s_delivery" refers to the average delivery time in the past six months. "td_o_success" refers to the number of successful deliveries in the past month. "td_t_success" refers to the number of successful deliveries in the past three months. "td_t_order" refers to the number of orders in the past three months. "td_t_delivery" refers to the average delivery time in the past three months. "td_s_success" refers to the number of successful deliveries in the past six months. The keys starting with "td" are corresponding features generated based on the material situation.

[0061] In this embodiment, the data fusion system 100 may be associated with the node 511 according to the input data to run the supplier selection mechanism. Next, the data fusion system 100 may point to the node 521 or the node 522 corresponding to the factor according to the input data of Table 3 below, and may further select one of the nodes 531 to 538 corresponding to the condition. Finally, the data fusion system 100 may point to the node 541 or the node 542 corresponding to the conclusion according to one of the nodes 531 to 538 to obtain the prediction result. In this embodiment, the prediction result may be, for example, displayed by the electronic device 210 operated by the user, as the conclusion information recorded in the node 541 or the node 542 (i.e., whether this supplier can be selected or not). Alternatively, in one embodiment, the prediction result may also include information such as the number of days of delay or the number of days of order delivery.

[0062]

[0063] Table 3

[0064] For another example, suppose a triple is: ([urgent, o_order, td_s_success], [urgent <= 0.383 and o_order <= 555.23 and td_s_success <= 2369.894 and td_s_success <= 202.942], 0.801). If the values ​​corresponding to the three features "urgent", "o_order" and "td_s_success" in a certain data satisfy the condition of this triple, the prediction result is that the delay probability is 0.801.

[0065] In summary, the data fusion system and data fusion method of the present invention can automatically construct a prediction model based on form data, and automatically integrate the data of the prediction model into the knowledge graph data executed by the system, so as to effectively reduce the redundancy and complexity of the knowledge graph data after amplification, and thereby enable the system to realize effective business prediction functions based on the knowledge graph data.

[0066] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A data fusion system, characterized in that: include: A storage device for storing a feature extraction module, a model construction module, a model disassembly module and a data fusion module; as well as a processor, electrically connected to the storage device, for executing the feature extraction module, the model construction module, the model disassembly module and the data fusion module, and for receiving a plurality of form data, wherein the feature extraction module extracts a plurality of feature data from the plurality of form data, the model construction module performs preprocessing operations on the plurality of feature data, and constructs a prediction model, The model decomposition module decomposes the prediction model into multiple triples (Triples Data), and the data fusion module fuses the multiple triples data into the knowledge graph data.

2. The data fusion system according to claim 1, characterized in that: The multiple form data include at least one of a historical purchase form, a purchase form, and a supplier basic information form.

3. The data fusion system according to claim 1, characterized in that: The feature extraction module performs table association processing on the plurality of form data to generate summary table data, and performs ETL (Extract, Transform, Load) engineering calculation on the summary table data to generate the plurality of feature data.

4. The data fusion system according to claim 1, characterized in that: The preprocessing operation includes performing data elimination and data filling on the plurality of feature data, selecting relevant features according to variance analysis to generate a plurality of sample data, and establishing a sample label pair set according to the plurality of sample data.

5. The data fusion system according to claim 4, characterized in that: The model building module builds the prediction model according to the sample label pair set and based on a decision tree algorithm.

6. The data fusion system according to claim 5, characterized in that: The prediction model is a tree structure, and each node data of the tree structure includes information entropy, sample quantity information, category quantity information and category information.

7. The data fusion system according to claim 1, characterized in that: The plurality of triplet data are respectively composed of factor data, condition data and conclusion data.

8. The data fusion system according to claim 1, characterized in that: The data fusion module fuses the multiple triple data into the knowledge graph data according to the set relationship between the factor data and the condition data of the multiple triple data and the data size relationship between the conclusion data of the multiple triple data.

9. The data fusion system according to claim 8, characterized in that: The data fusion module merges at least a portion of the multiple triple data and fuses them into the knowledge graph data.

10. The data fusion system according to claim 1, characterized in that: The processor receives input data, and searches the knowledge graph data according to the input data to generate a prediction result.

11. A data fusion method, characterized in that: include: Executing a feature extraction module through a processor to extract a plurality of feature data from a plurality of form data; Executing a model building module by the processor to perform preprocessing operations on the plurality of feature data and build a prediction model; Executing a model disassembly module by the processor to disassemble the prediction model into a plurality of triple data; as well as The data fusion module is executed by the processor to fuse the multiple triple data into the knowledge graph data.

12. The data fusion method according to claim 11, characterized in that: The multiple form data include at least one of a historical purchase form, a purchase form, and a supplier basic information form.

13. The data fusion method according to claim 11, characterized in that: The step of extracting the plurality of feature data from the plurality of form data comprises: Performing table association processing on the plurality of form data to generate summary table data; and An ETL (Extract, Transform, Load) engineering calculation is performed on the summary table data to generate the plurality of feature data.

14. The data fusion method according to claim 11, characterized in that: The pre-processing operation includes: Performing data elimination and data filling on the plurality of feature data, and selecting relevant features according to variance analysis to generate a plurality of sample data; and A sample label pair set is established according to the plurality of sample data.

15. The data fusion method according to claim 14, characterized in that: The steps of constructing the prediction model include: The prediction model is constructed according to the sample label pair set and based on a decision tree algorithm.

16. The data fusion method according to claim 15, characterized in that: The prediction model is a tree structure, and each node data of the tree structure includes information entropy, sample quantity information, category quantity information and category information.

17. The data fusion method according to claim 11, characterized in that: The plurality of triplet data are respectively composed of factor data, condition data and conclusion data.

18. The data fusion method according to claim 11, characterized in that: The data fusion module fuses the multiple triple data into the knowledge graph data according to the set relationship between the factor data and the condition data of the multiple triple data and the data size relationship between the conclusion data of the multiple triple data.

19. The data fusion method according to claim 18, characterized in that: The step of fusing the plurality of triple data into the knowledge graph data comprises: At least a portion of the multiple triple data are merged and integrated into the knowledge graph data.

20. The data fusion method according to claim 11, characterized in that: Also includes: receiving input data by the processor; as well as The processor searches the knowledge graph data according to the input data to generate a prediction result.

Images