Knowledge graph updating method introducing time sequence analysis and incremental learning algorithm

Abstract

The invention belongs to the technical field of knowledge maps, and discloses a knowledge map updating method introducing a time sequence analysis and incremental learning algorithm. Comprising the following steps: S1, collecting new cases of an enterprise, and performing knowledge extraction on the new cases to form a new case sub-graph; s2, performing time sequence analysis on the new case sub-graph to obtain a development trend of a new case; according to the method, the new case sub-graph can be conveniently fused with the sub-graphs of the same category of products with the same development trend in the original knowledge graph, and an enterprise can conveniently formulate a subsequent sales strategy and a product optimization strategy of the products corresponding to the new case sub-graph; by recording the resolution path and setting the jump burying point, the fusion processing time of the new case sub-graph and the original knowledge graph can be greatly shortened, and the updating efficiency and stability of the original knowledge graph are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of knowledge graph, more particularly, the present application relates to a knowledge graph updating method introducing time series analysis and incremental learning algorithm. BACKGROUND

[0002] In the process of enterprise operation and decision-making, knowledge graph, as a powerful knowledge representation and management tool, plays a crucial role; it can integrate multi-source data inside and outside the enterprise, build a complex knowledge network, and provide comprehensive and accurate information support for enterprises to make scientific and reasonable decisions; with the continuous change of market environment and the continuous expansion of enterprise business, new cases (such as electronic products) of enterprises are emerging, which contain new product information, market dynamics and user demand and other key knowledge, and the knowledge in these new cases needs to be integrated into the original knowledge graph to update the original knowledge graph.

[0003] However, when the new case subgraph corresponding to the new case is fused with the original knowledge graph, the following two key problems are faced: first, in the case of a variety of enterprise products, how to accurately fuse the most matched subgraph of the new case subgraph with the original knowledge graph; second, conflict nodes are prone to occur in the fusion process of the new case subgraph and the original knowledge graph, how to quickly eliminate these conflict nodes.

[0004] In view of this, the present application proposes a knowledge graph updating method introducing time series analysis and incremental learning algorithm to solve the above problems. SUMMARY

[0005] In order to overcome the above-mentioned defects of the prior art, in order to achieve the above-mentioned purpose, the present application provides the following technical scheme: a knowledge graph updating method introducing time series analysis and incremental learning algorithm, comprising the following steps:

[0006] S1, collecting new cases of an enterprise, and performing knowledge extraction on the new cases to form a new case subgraph;

[0007] S2, performing time series analysis on the new case subgraph to obtain the development trend of the new case;

[0008] S3, fusing the new case subgraph with the original knowledge graph according to the development trend, if it can be completely fused, the original knowledge graph updating is completed, if it cannot be completely fused, marking the nodes in the new case subgraph that cannot be fused as conflict nodes;

[0009] S4, performing conflict resolution on the conflict nodes according to the preset rules until the original knowledge graph updating is completed;

[0010] S5. Record the conflict resolution process to generate a resolution path, set jump tracking points on the resolution path, and use the jump tracking points to quickly resolve the conflict nodes of subsequent new cases until the subsequent new cases are updated in the original knowledge graph.

[0011] Furthermore, the step S1 includes:

[0012] S10. Collect new cases and new case data of enterprises from multiple channels, and perform knowledge extraction on the new case data to obtain products, product impact factors, users, and user impact factors corresponding to the new cases;

[0013] S11. Take the product, product influence factor, user and user influence factor corresponding to the new case as nodes, determine the relationship between nodes and connect the nodes with the corresponding relationship to obtain a new case subgraph, and the nodes are associated with corresponding timestamps and sales data corresponding to this timestamp.

[0014] Furthermore, the step S10 includes:

[0015] S100: Use the product name and model as product keywords, use the product's corresponding specification information and function information as product influencing factor keywords, use the user type of the product's corresponding user as the user keyword, and use the influencing factors corresponding to the user type as the user influencing factor keywords;

[0016] S101. Use natural language processing technology to extract product keywords, product influencing factor keywords, user keywords, and user influencing factor keywords, that is, obtain keywords of new cases, and classify and associate the keywords of new cases with corresponding product, product influencing factor, user, and user influencing factor categories.

[0017] Furthermore, the step S2 includes:

[0018] S20, aligning the sales data of each node based on the timestamp associated with the node, with time as the horizontal axis and sales data as the vertical axis, thereby establishing a sales trend graph of the new case subgraph, wherein the sales trend graph corresponds to a sales trend curve;

[0019] S21. Set a sliding window of n time units on the sales trend curve of the sales trend graph corresponding to the new case subgraph. Calculate the average sales data using the sliding window as a unit. Slide the window backward in chronological order, sliding 1 time unit each time, to obtain the development trend of the new case subgraph:

[0020] If the growth rate of the average sales data of the subsequent sliding window compared to the average sales data of the previous sliding window for m consecutive sliding windows is greater than the set threshold α, it is an upward trend;

[0021] If the average sales data of the latter sliding window of the consecutive m sliding windows is greater than the set threshold value a compared with the falling rate of the average sales data of the former sliding window, it is a falling trend;

[0022] If the average sales data of the consecutive m sliding windows presents an up-down alternating fluctuation state, and the absolute value of the growth rate or the falling rate of the average sales data of adjacent sliding windows is greater than the set threshold value γ, it is an oscillation trend;

[0023] If the fluctuation amplitude of the average sales data of the consecutive m sliding windows is less than the set threshold value δ, that is, the difference between the average sales data of each sliding window is small, it is a stable trend.

[0024] Further, in the S20,

[0025] S200, determine the wave peak and wave trough in the sales trend graph corresponding to the new case subgraph, and determine the time stamp corresponding to the wave peak and wave trough, wherein the time stamp corresponding to the wave peak is taken as the wave peak point, and the time stamp corresponding to the wave trough is taken as the wave trough point;

[0026] S201, the wave peak point corresponding time stamp until the wave peak range corresponding to the previous time stamp is the first boundary, and the rising observation area is obtained, and the wave trough point corresponding time stamp until the wave trough range corresponding to the previous time stamp is the second boundary, and the falling observation area is obtained;

[0027] S202, extract the newly added or reduced nodes in the rising observation area and the appearance frequency of each node, and perform clustering processing on the newly added or reduced nodes in the rising observation area and the appearance frequency of each node, to obtain the rising keywords, extract the newly added or reduced nodes in the falling observation area and the appearance frequency of each node, and perform clustering processing on the newly added or reduced nodes in the falling observation area and the appearance frequency of each node, to obtain the falling keywords;

[0028] S203, associate the rising keywords or the falling keywords with the development trend of the corresponding new case subgraph.

[0029] Further, the step of S3 comprises:

[0030] S30, obtaining other case subgraphs of the same category as the product corresponding to the new case subgraph in the original knowledge graph as the to-be-fused subgraph;

[0031] S31, taking the time between the starting time stamp and the current time stamp corresponding to the new case subgraph as the basic time period, and obtaining the sales trend curve of the to-be-fused subgraph in the adjacent basic time period as the to-be-compared curve;

[0032] S32, obtain a matching degree of the sales trend curve of the new case subgraph and the to-be-compared curve, and take the to-be-fused subgraph corresponding to the to-be-compared curve with the highest matching degree as the fusible subgraph;

[0033] S33, summarize the keywords of the nodes in the new case subgraph into a first set, summarize the keywords of the nodes in the fusible subgraph into a second set, merge the nodes with the same keywords in the first set and the second set, and take the nodes with different keywords in the first set and the second set as conflict nodes.

[0034] Further, the step S32 comprises:

[0035] S320, superimpose the sales trend curve of the new case subgraph and the to-be-compared curve in a virtual grid, mark the number of superimposed grids as Q1, mark the number of grids occupied by the to-be-compared curve in the virtual grid as Q, and calculate the ratio of Q1 to Q, that is, obtain the matching degree of the sales trend curve of the new case subgraph and the to-be-compared curve.

[0036] Further, the step S4 comprises:

[0037] S40, preset rules, the rules including replacing the keywords of the conflict nodes, decomposing the keywords of the conflict nodes, and extending the keywords of the conflict nodes, each rule corresponding to a rule number;

[0038] S41, obtain the synonyms of the keywords corresponding to the conflict nodes based on the word vectors, if the synonyms corresponding to the conflict nodes are the same as the keywords of the nodes in the second set, replace the keywords of the conflict nodes with the synonyms corresponding to the conflict nodes and merge the conflict nodes with the nodes in the second set with the same keywords;

[0039] S42, decompose the keywords of the conflict nodes into multiple lower-level words by using semantic analysis, if one of the multiple lower-level words is the same as the keywords of the nodes in the second set, merge the conflict nodes with the nodes in the second set with the same keywords, and the remaining lower-level words are used as new nodes to connect the nodes corresponding to the same lower-level words;

[0040] S42, obtain the superordinate words, bridge words, reverse constraint words, and special scene words of the conflict nodes as the keywords of the extended conflict nodes, and merge the conflict nodes with the nodes in the second set by using the keywords of the extended conflict nodes;

[0041] S42, perform conflict resolution on the conflict nodes of the new case subgraph according to the above steps, until all the conflict nodes are fused with the nodes in the second set, that is, the updating of the original knowledge graph is completed.

[0042] Further, the step S5 comprises:

[0043] S50, record each conflict resolution step, each step as a path node and connect in order to form a resolution path, the path node corresponds to a rule number;

[0044] S51, the starting path node of the resolution path is taken as a jump buried point, and a jump path is established between the jump buried point and the final path node;

[0045] S52, obtain the conflict difference value between the conflict node of the subsequent new case and the starting path node of the resolution path, if the conflict difference value meets the preset threshold interval, jump the conflict node of the new case to the corresponding final path node through the jump buried point to complete the resolution of the new case conflict node, until the update of the subsequent new case in the original knowledge graph is completed.

[0046] Further, the step S52 comprises:

[0047] S520, set the keyword corresponding to the conflict node of the subsequent new case as A, and set the keyword corresponding to the starting path node of the resolution path as B;

[0048] S521, calculate the conflict difference value of the keyword A and the keyword B, and the expression is:

[0049] CTZ=ω1xCJL+(1-ω1)xCXS;

[0050] In the formula, CTZ represents the conflict difference value between the conflict node of the subsequent new case and the starting path node of the resolution path, ω1 represents the weight coefficient corresponding to CJL, CJL represents the normalized edit distance of the keyword A and the keyword B, and CXS represents the semantic similarity of the keyword A and the keyword B.

[0051] The technical effects and advantages of the knowledge graph updating method introduced by the application of time series analysis and incremental learning algorithm are:

[0052] 1. By judging the development trend of the new case subgraph, the same category product subgraph with the same development trend in the original knowledge graph can be easily fused, and the product of the subsequent new case subgraph can be taken as a reference to develop the same category product as a reference, so that the enterprise can develop the subsequent sales strategy and product optimization strategy of the new case subgraph corresponding product, and provide a good development direction for the new case subgraph corresponding product;

[0053] 2. By recording the resolution path and setting the jump buried point, the historical conflict resolution experience can be converted into reusable jump path, so as to facilitate the fusion of the new case subgraph and the original knowledge graph, which can greatly shorten the fusion processing time of the new case subgraph and the original knowledge graph, and improve the update efficiency and stability of the original knowledge graph. BRIEF DESCRIPTION OF DRAWINGS

[0054] Figure 1 This is a flow chart of the knowledge graph updating method that introduces time series analysis and incremental learning algorithm of the present invention. DETAILED DESCRIPTION

[0055] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0056] See also Figure 1 As shown, the knowledge graph updating method of this embodiment, which introduces time series analysis and incremental learning algorithm, includes the following steps:

[0057] S1. Collect new cases of enterprises and extract knowledge from them to form new case subgraphs;

[0058] S2. Perform time series analysis on the new case subgraph to obtain the development trend of the new case;

[0059] S3. Integrate the new case subgraph with the original knowledge graph according to the development trend. If the new case subgraph can be fully integrated, the original knowledge graph is updated. If the new case subgraph cannot be fully integrated, the nodes that cannot be integrated in the new case subgraph are marked as conflict nodes.

[0060] S4. Resolve conflicts among conflicting nodes according to preset rules until the original knowledge graph is updated;

[0061] S5. Record the conflict resolution process to generate a resolution path. Set jump tracking points on the resolution path. Use the jump tracking points to quickly resolve conflict nodes in subsequent new cases until the subsequent new cases are updated in the original knowledge graph.

[0062] In this embodiment, by judging the development trend of the new case subgraph, it is possible to facilitate the fusion of the new case subgraph with the subgraph of the same category of products with the same development trend in the original knowledge graph, so that subsequent products of the new case subgraph can refer to the same category of products with the same development trend as a reference, so that the company can formulate subsequent sales strategies and product optimization strategies for the products corresponding to the new case subgraph; by recording the resolution path and setting jump points, the historical conflict resolution experience can be converted into a reusable jump path to facilitate the fusion of the new case subgraph with the original knowledge graph, which can greatly shorten the fusion processing time of the new case subgraph and the original knowledge graph, and improve the update efficiency and stability of the original knowledge graph.

[0063] As an optional embodiment: the step S1 includes:

[0064] S10, collect new cases and new case data of the enterprise from multiple channels, and perform knowledge extraction on the new case data to obtain the product, product influence factor, user, and user influence factor corresponding to the new case;

[0065] S11, taking the product, product influence factor, user, and user influence factor corresponding to the new case as nodes, determining the relationship between the nodes and connecting the nodes according to the corresponding relationship to obtain a new case subgraph, the nodes being associated with a corresponding timestamp and sales data corresponding to the timestamp;

[0066] It should be further noted that the new case of the enterprise is a new product (such as an electronic product) of the enterprise, and the new case data includes but is not limited to that obtained from multiple channels such as official websites, news reports, industry analysis reports, social media platforms, and user feedback. The product influence factor is, for example, a function parameter (battery capacity, processor model), a size specification (screen size, body thickness), and the like. The user factor is, for example, a user type (such as an ordinary consumer, a professional user, and an enterprise customer), a purchase motivation (such as cost performance and brand preference), a use scenario (such as daily use, business office, and game entertainment), and the like. The relationship between the product and the product influence factor (such as a W phone having an R processor); the relationship between the user and the user influence factor (such as user A preferring a high cost performance product); the relationship between the user and the product (such as user A purchasing a W phone); the relationship between the timestamp and the sales data (such as 1000 W phones being sold in January 2024); and further, by collecting and extracting new case data from multiple channels, the product, user, and related influence factors can be accurately extracted to form structured knowledge nodes, thereby providing comprehensive data support for product sales and development decisions.

[0067] As an optional embodiment, the step S10 includes:

[0068] S100, taking the name and model number of the product as the key words of the product, taking the specification information and function information corresponding to the product as the key words of the product influence factor, taking the user type of the user corresponding to the product as the key words of the user, and taking the influence factor corresponding to the user type as the key words of the user influence factor;

[0069] S101, using natural language processing technology (such as NLP) to extract the key words of the product, the key words of the product influence factor, the key words of the user, and the key words of the user influence factor, i.e., to obtain the key words of the new case, and classify and associate the key words of the new case to the corresponding product, product influence factor, user, and user influence factor category;

[0070] It also needs to be explained that the preprocessing of new case data includes marking, word segmentation and part of speech, and removing noise information, including web page tags, special symbols, etc. The process of extracting keywords using natural language processing technology (such as NLP) is as follows: extracting from news reports: "W mobile phone R processor performance is strong, suitable for students who enjoy gaming"; extracting keywords: W mobile phone (product), R processor (product influence factor), gaming (user influence factor); students who enjoy gaming (users); classification: "W mobile phone" is classified into products, "R processor" is classified into product influence factors, "gaming" is classified into user influence factors, and students are classified into users. Users can also be modified as gaming enthusiasts; Through the above knowledge extraction and new case subgraph construction, enterprises can accurately extract products, users and related influence factors, and form structured nodes; By constructing a new case subgraph based on node relationships, combining time stamps and sales data, market dynamics and user demand changes can be intuitively presented, helping enterprises accurately grasp the development trend of products, optimize product strategies, and provide a basis for product sales strategies; Further, the new case subgraph can effectively integrate scattered information and provide direction for intelligent decision-making and operation of enterprises.

[0071] As an optional embodiment, the step S2 comprises:

[0072] S20, aligning the sales data of each node based on the time stamp of the node association, taking time as the horizontal axis and sales data as the vertical axis, to establish a sales trend graph of the new case subgraph, and the sales trend graph corresponds to a sales trend curve;

[0073] S21, setting a sliding window with a size of n time units (such as n days, n months) on the sales trend curve of the new case subgraph corresponding sales trend graph, calculating the average sales data in units of sliding window, and sliding the window backward in time order each time by 1 time unit, to obtain the development trend of the new case subgraph:

[0074] If the growth rate of the average sales data of the next sliding window compared with the average sales data of the previous sliding window in the continuous m sliding windows is greater than the set threshold value a, it is an upward trend;

[0075] If the decline rate of the average sales data of the next sliding window compared with the average sales data of the previous sliding window in the continuous m sliding windows is greater than the set threshold value a, it is a downward trend;

[0076] If the average sales data of the continuous m sliding windows presents an alternating fluctuation state of rising and falling, and the absolute values of the growth rate or the decline rate of the average sales data of adjacent sliding windows are greater than the set threshold value γ, it is an oscillation trend;

[0077] If the fluctuation range of the average sales data of the continuous m sliding windows is less than the set threshold value δ, that is, the difference between the average sales data of each sliding window is small, it is a stable trend;

[0078] It should be further pointed out that the threshold value α, the threshold value γ and the threshold value δ are set by relevant personnel according to the trend of historical sales data. For example, when the average sales growth rate of the continuous 3 sliding windows (1-3 months, 2-4 months, 3-5 months) is more than 5%, it is determined as an upward trend; for example, when the average sales decrease rate of the continuous 3 sliding windows is more than 5%, it is determined as a downward trend; for example, among the 10 continuous sliding windows, the average sales appears "increase-decrease-increase-decrease" alternately, and the change amplitude is more than 3% each time, it is determined as an oscillation trend; for example, the fluctuation range of the average sales of the continuous 10 sliding windows is within 2%, it is determined as a stable trend; further, through the above judgment of the development trend of the new case subgraph, the new case subgraph can be easily merged with the subgraph of the same category product with the same development trend in the original knowledge graph, and the product of the subsequent new case subgraph can be easily referred to as a reference with the same development trend of the same category product, so as to facilitate the enterprise to formulate the subsequent sales strategy and product optimization strategy of the product corresponding to the new case subgraph, and provide a good development direction for the product corresponding to the new case subgraph.

[0079] As an optional embodiment, in the S20,

[0080] S200, determining the wave crest and the wave trough in the sales trend graph corresponding to the new case subgraph, and determining the time stamp corresponding to the wave crest and the wave trough, wherein the time stamp corresponding to the wave crest is taken as the wave crest point, and the time stamp corresponding to the wave trough is taken as the wave trough point;

[0081] S201, taking the wave crest point corresponding time stamp until the previous time stamp corresponding to the wave crest range as the first boundary to obtain the rising observation area, and taking the wave trough point corresponding time stamp until the previous time stamp corresponding to the wave trough range as the second boundary to obtain the falling observation area;

[0082] S202, extracting the newly added or reduced nodes in the rising observation area and the appearance frequency of each node, and performing clustering processing on the newly added or reduced nodes in the rising observation area and the appearance frequency of each node to obtain the rising keyword, and extracting the newly added or reduced nodes in the falling observation area and the appearance frequency of each node, and performing clustering processing on the newly added or reduced nodes in the falling observation area and the appearance frequency of each node to obtain the falling keyword;

[0083] S203, associating the rising keyword or the falling keyword with the development trend of the corresponding new case subgraph;

[0084] It also needs to be explained that through the acquisition of the rising keywords and the falling keywords, the maximum influencing factor affecting the rising or falling of the sales data can be obtained, so that the enterprise can improve the product or the marketing of the product according to the maximum influencing factor, so as to obtain greater sales data or correct the defects of the product, and then the enterprise can formulate the subsequent sales strategy and product optimization strategy of the new case subgraph corresponding product, and provide a good development direction for the new case subgraph corresponding product.

[0085] As an optional embodiment: the step of S3 comprises:

[0086] S30, acquiring other case subgraphs of the same category as the product corresponding to the new case subgraph in the original knowledge graph as a to-be-fused subgraph;

[0087] S31, taking the time between the starting timestamp of the new case subgraph and the current timestamp as the basic time period, and acquiring the sales trend curve of the to-be-fused subgraph in the adjacent basic time period as the to-be-compared curve;

[0088] S32, acquiring the matching degree of the sales trend curve of the new case subgraph and the to-be-compared curve, and taking the to-be-fused subgraph corresponding to the to-be-compared curve with the highest matching degree as a fusable subgraph;

[0089] S33, summarizing the keywords of the nodes in the new case subgraph into a first set, summarizing the keywords of the nodes in the fusable subgraph into a second set, merging the nodes with the same keywords in the first set and the second set, and taking the nodes with different keywords in the first set and the second set as conflict nodes;

[0090] It also needs to be explained that the to-be-fused subgraph with the highest matching degree can be screened as a fusable subgraph, so that the new case subgraph can be fused with the original knowledge graph, and such fusion can make the new case subgraph take the to-be-fused subgraph as a reference in subsequent product optimization or sales strategy, so as to formulate the subsequent development strategy of the new case subgraph corresponding product.

[0091] As an optional embodiment: the step of S32 comprises:

[0092] S320, superimposing the sales trend curve of the new case subgraph and the to-be-compared curve in a virtual grid, marking the number of superimposed grids as Q1, marking the number of grids occupied by the to-be-compared curve in the virtual grid as Q, and calculating the ratio of Q1 and Q, that is, obtaining the matching degree of the sales trend curve of the new case subgraph and the to-be-compared curve;

[0093] It should be further explained that the higher the matching degree is, the more the number of grids in which the sales trend curve of the new case subgraph coincides with the to-be-compared curve, i.e., the higher the similarity between the two is; the lower the matching degree is, the fewer the number of grids in which the sales trend curve of the new case subgraph coincides with the to-be-compared curve, i.e., the lower the similarity between the two is.

[0094] As an optional embodiment, the step of S4 comprises:

[0095] S40, preset rules, the rules comprising keywords for replacing conflict nodes, keywords for decomposing conflict nodes and keywords for extending conflict nodes, each rule corresponding to a rule number;

[0096] S41, based on a word vector (such as Word2Vec or BERT), a synonym of a keyword corresponding to a conflict node is obtained, if the synonym corresponding to the conflict node is the same as a keyword of a node in the second set, the synonym corresponding to the conflict node is replaced with the keyword of the conflict node and the node with the same keyword in the second set is merged;

[0097] S42, the keyword of the conflict node is decomposed into multiple lower-level words by using semantic analysis, if one of the multiple lower-level words is the same as a keyword of a node in the second set, the conflict node and the node with the same keyword in the second set are merged, and the remaining lower-level words are used as new nodes and connected to the nodes with the same lower-level words;

[0098] S42, the upper-level words, bridge words, reverse constraint words and special scene words of the conflict node are obtained as the keywords of the extended conflict node, and the conflict node and the nodes in the second set are merged by using the keywords of the extended conflict node;

[0099] S42, the conflict nodes of the new case subgraph are resolved according to the above steps until all the conflict nodes are fused with the nodes in the second set, i.e., the updating of the original knowledge graph is completed;

[0100] It should also be noted that the keywords of the replacement conflict nodes, such as the conflict node "mobile phone", are replaced with the standard word "smartphone"; for example, "product qualification rate" and "yield rate" are uniformly replaced with "product qualification rate"; the keywords of the decomposed conflict nodes, such as "smartphone", are decomposed into "touch screen mobile phone" and "folding screen mobile phone"; the keywords of the extended conflict nodes, such as the conflict node "smart watch", have no direct connection with the original node "health monitoring" → insert the bridging word "heart rate monitoring function", and the application store is used as a bridging word to connect "smartphone" and "software application"; hypernyms such as "smart device" are used as the hypernym of "smart watch", reverse constraint words: non-smartphones (excluding feature phones), special scenario words: waterproof mobile phones (suitable for outdoor scenes), elderly phones (suitable for specific user groups); thus, through the above preset rules, the conflict nodes can be quickly resolved to facilitate the integration of the new case subgraph with the original knowledge graph, and to facilitate the update of the original knowledge graph.

[0101] As an optional embodiment: the step S5 includes:

[0102] S50, recording each conflict resolution step, taking each step as a path node and connecting them in order to form a resolution path, wherein the path nodes correspond to rule numbers;

[0103] S51: Use the starting path node of the resolved path as a jump embedding point, and establish a jump path between the jump embedding point and the final path node;

[0104] S52: Obtain the conflict difference between the conflict node of the subsequent new case and the starting path node of the resolution path. If the conflict difference satisfies a preset threshold interval, jump the conflict node of the new case to the corresponding final path node through a jump embedding point to complete the resolution of the conflict node of the new case, until the subsequent new case is updated in the original knowledge graph;

[0105] It should also be noted that the initial resolution step of the resolution path is taken as the starting path node, the starting path node corresponds to the conflict node, and the final resolution step of the resolution path is taken as the final path node, the final path node corresponds to the node in the second set merged by the conflict node; the threshold interval is obtained by ±3σ of the maximum conflict difference between the keyword corresponding to the starting path node and the corresponding multiple synonyms; by recording the resolution path and setting jump points, historical conflict resolution experience can be converted into a reusable jump path, which is especially suitable for high-frequency, low-complexity conflict scenarios (such as sales strategy formulation for electronic products, optimization and development strategy formulation for electronic products, etc.). Furthermore, the jump path can greatly shorten the fusion processing time of the new case subgraph and the original knowledge graph while ensuring the accuracy of the resolution, thereby improving the update efficiency and stability of the original knowledge graph.

[0106] As an optional embodiment: the step of S52 includes:

[0107] S520, set the keyword corresponding to the conflict node of the subsequent new case as A, and set the keyword corresponding to the starting path node of the resolution path as B;

[0108] S521, calculate the conflict difference value of the keyword A and the keyword B, and the expression is:

[0109] CTZ = ω1 x CJL + (1- ω1) x CXS;

[0110] In the formula, CTZ represents the conflict difference value of the conflict node of the subsequent new case and the starting path node of the resolution path, ω1 represents the weight coefficient corresponding to CJL, CJL represents the normalized edit distance of the keyword A and the keyword B, 1- ω1 represents the weight coefficient of CXS, and CXS represents the semantic similarity of the keyword A and the keyword B;

[0111] It should be further pointed out that ω1 is set by relevant personnel according to the importance of the semantic similarity or the normalized edit distance in the actual resolution path, For example, set two keywords "apple" and "apples", the edit distance (apple", "apples") = 1, the normalized edit distance = 1 / max(5, 6) = 1 / 6 ≈ 0.167, and the semantic similarity of "apple" and "apples" is calculated using the BERT word embedding model, V apple = [0.1, 0.2, 0.3…], V apples = [0.11, 0.21, 0.31…], Further, the semantic similarity and the normalized edit distance can reflect the difference between the keyword corresponding to the conflict node of the subsequent new case and the starting path node of the resolution path, and can facilitate finding the starting path node with the smallest conflict difference value as the resolution path based on the difference between the two, to assist the subsequent new case in finding the starting path node with the smallest conflict difference value as the resolution path to realize the quick resolution of the conflict node corresponding to the subsequent new case, thereby improving the fusion speed of the subsequent new case and the original knowledge graph, and improving the update speed of the original knowledge graph.

[0112] Those skilled in the art can realize that the units and algorithm steps of each example described in combination with the embodiments disclosed in the present application can be realized by electronic hardware or a combination of computer software and electronic hardware. Whether the functions are realized in hardware or software depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods to realize the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0113] In several embodiments provided by the present application, it should be understood that the disclosed system, device and method can be implemented in other ways. For example, the device embodiments described above are merely illustrative, and the division of the units is only one of the ways, and in actual implementation, other division ways can be used, for example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the units shown or discussed can be indirect coupling or communication connection through some interfaces, devices or units, which can be electrical, mechanical or other forms.

[0114] The above description is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto, and any person skilled in the art can easily think of changes or replacements within the technical range disclosed by the present application, which should be included in the protection scope of the present application.

[0115] Finally: the above description is only the preferred embodiment of the present application and is not used to limit the present application, and any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included in the protection scope of the present application.

Claims

1. A knowledge graph updating method that introduces time series analysis and incremental learning algorithm, characterized in that: The following steps are involved: S1. Collect new cases of enterprises and extract knowledge from them to form new case subgraphs; S2. Perform time series analysis on the new case subgraph to obtain the development trend of the new case; S3. Integrate the new case subgraph with the original knowledge graph according to the development trend. If the new case subgraph can be fully integrated, the original knowledge graph is updated. If the new case subgraph cannot be fully integrated, the nodes that cannot be integrated in the new case subgraph are marked as conflict nodes. S4. Resolve conflicts among conflicting nodes according to preset rules until the original knowledge graph is updated; S5. Record the conflict resolution process to generate a resolution path, set jump tracking points on the resolution path, and use the jump tracking points to quickly resolve the conflict nodes of subsequent new cases until the subsequent new cases are updated in the original knowledge graph.

2. The knowledge graph updating method introducing time series analysis and incremental learning algorithm according to claim 1 is characterized in that: The steps of S1 include: S10. Collect new cases and new case data of enterprises from multiple channels, and perform knowledge extraction on the new case data to obtain products, product impact factors, users, and user impact factors corresponding to the new cases; S11. Take the product, product influence factor, user and user influence factor corresponding to the new case as nodes, determine the relationship between nodes and connect the nodes with the corresponding relationship to obtain a new case subgraph, and the nodes are associated with corresponding timestamps and sales data corresponding to this timestamp.

3. The knowledge graph updating method according to claim 2, characterized in that: The step of S10 includes: S100: Use the product name and model as product keywords, use the product's corresponding specification information and function information as product influencing factor keywords, use the user type of the product's corresponding user as the user keyword, and use the influencing factors corresponding to the user type as the user influencing factor keywords; S101. Use natural language processing technology to extract product keywords, product influencing factor keywords, user keywords, and user influencing factor keywords, that is, obtain keywords of new cases, and classify and associate the keywords of new cases with corresponding product, product influencing factor, user, and user influencing factor categories.

4. The knowledge graph updating method according to claim 3, characterized in that: The steps of S2 include: S20, aligning the sales data of each node based on the timestamp associated with the node, with time as the horizontal axis and sales data as the vertical axis, thereby establishing a sales trend graph of the new case subgraph, wherein the sales trend graph corresponds to a sales trend curve; S21. Set a sliding window of n time units on the sales trend curve of the sales trend graph corresponding to the new case subgraph. Calculate the average sales data using the sliding window as a unit. Slide the window backward in chronological order, sliding 1 time unit each time, to obtain the development trend of the new case subgraph: If the growth rate of the average sales data of the subsequent sliding window compared to the average sales data of the previous sliding window for m consecutive sliding windows is greater than the set threshold α, it is an upward trend; If the decline rate of the average sales data of the subsequent sliding window compared to the average sales data of the previous sliding window for m consecutive sliding windows is greater than the set threshold α, it is a downward trend; If the average sales data of m consecutive sliding windows show an alternating upward-downward fluctuation state, and the absolute value of the growth rate or decline rate of the average sales data of adjacent sliding windows is greater than the set threshold γ, then it is an oscillation trend; If the fluctuation range of the average sales data of m consecutive sliding windows is less than the set threshold δ, that is, the difference between the average sales data of each sliding window is small, it is a stable trend.

5. The knowledge graph updating method according to claim 4, characterized in that: In said S20, S200, determining the peaks and troughs in the sales trend graph corresponding to the new case subgraph, and determining the timestamps corresponding to the peaks and troughs, wherein the timestamps corresponding to the peaks are used as peak points, and the timestamps corresponding to the troughs are used as trough points; S201: The timestamp corresponding to the peak point to the previous timestamp corresponding to the peak range is used as the first boundary to obtain a rising observation area, and the timestamp corresponding to the trough point to the previous timestamp corresponding to the trough range is used as the second boundary to obtain a falling observation area; S202, extracting nodes that are newly added or reduced in the rising observation area and the frequency of occurrence of each node, performing clustering processing on the nodes that are newly added or reduced in the rising observation area and the frequency of occurrence of each node, thereby obtaining rising keywords; extracting nodes that are newly added or reduced in the falling observation area and the frequency of occurrence of each node, performing clustering processing on the nodes that are newly added or reduced in the falling observation area and the frequency of occurrence of each node, thereby obtaining falling keywords; S203: Associating the rising keywords or the falling keywords with the development trend of the corresponding new case subgraph.

6. The knowledge graph updating method according to claim 5, characterized in that: The steps of S3 include: S30, obtaining other case subgraphs in the original knowledge graph whose products belong to the same category as the new case subgraph as the subgraph to be fused; S31, using the time between the starting timestamp and the current timestamp corresponding to the new case subgraph as the basic time period, obtaining the sales trend curve of the subgraph to be fused within the adjacent basic time period as the curve to be compared; S32. Obtain the matching degree between the sales trend curve of the new case subgraph and the curve to be compared, and select the subgraph to be fused corresponding to the curve to be compared with the highest matching degree as the fusionable subgraph; S33. Aggregate the keywords of the nodes in the new case subgraph into a first set, aggregate the keywords of the nodes in the fusible subgraph into a second set, merge the nodes with the same keywords in the first set and the second set, and treat the nodes with different keywords in the first set and the second set as conflicting nodes.

7. The knowledge graph updating method according to claim 6, characterized in that: The steps of S32 include: S320. Overlap the sales trend curve of the new case subgraph with the curve to be compared in the virtual grid, mark the number of overlapping grids as Q1, mark the number of grids occupied by the curve to be compared in the virtual grid as Q, calculate the ratio of Q1 to Q, and obtain the matching degree between the sales trend curve of the new case subgraph and the curve to be compared.

8. The knowledge graph updating method according to claim 7, characterized in that: The steps of S4 include: S40, presetting rules, wherein the rules include replacing keywords of conflicting nodes, decomposing keywords of conflicting nodes, and extending keywords of conflicting nodes, and each rule has a corresponding rule number; S41. Obtain synonyms of the keywords corresponding to the conflicting node based on the word vector. If the synonyms corresponding to the conflicting node are the same as the keywords of the nodes in the second set, replace the keywords of the conflicting node with the synonyms corresponding to the conflicting node and merge the nodes with the same keywords in the second set. S42. Decompose the keyword of the conflicting node into multiple hyponyms using semantic parsing. If one of the multiple hyponyms is the same as the keyword of a node in the second set, merge the conflicting node with the node with the same keyword in the second set, and connect the remaining hyponyms as new nodes to the node with the same hyponym. S42: obtaining hypernyms, bridging words, reverse constraint words, and special scenario words of the conflicting node as keywords for extending the conflicting node, and merging the conflicting node with the nodes in the second set using the keywords of the extended conflicting node; S42. Resolve conflicts on the conflicting nodes of the new case subgraph according to the above steps until all conflicting nodes are merged with the nodes in the second set, thus completing the update of the original knowledge graph.

9. The knowledge graph updating method introducing time series analysis and incremental learning algorithm according to claim 1 is characterized in that: The steps of S5 include: S50, recording each conflict resolution step, taking each step as a path node and connecting them in order to form a resolution path, wherein the path nodes correspond to rule numbers; S51: Use the starting path node of the resolved path as a jump embedding point, and establish a jump path between the jump embedding point and the final path node; S52. Obtain the conflict difference between the conflict node of the subsequent new case and the starting path node of the resolution path. If the conflict difference meets the preset threshold interval, the conflict node of the new case is jumped to the corresponding final path node through the jump point to complete the resolution of the new case conflict node, until the subsequent new case is updated in the original knowledge graph.

10. The knowledge graph updating method according to claim 9, characterized in that: The steps of S52 include: S520: Set the keyword corresponding to the conflict node of the subsequent new case to A, and set the keyword corresponding to the starting path node of the resolution path to B; S521. Calculate the conflict difference between keyword A and keyword B. The expression is: CTZ=ω1×CJL+(1-ω1)×CXS; Where CTZ represents the conflict difference between the conflict node of the subsequent new case and the starting path node of the resolution path, ω1 represents the weight coefficient corresponding to CJL, CJL represents the normalized edit distance between keyword A and keyword B, and CXS represents the semantic similarity between keyword A and keyword B.

Images