Financial message pushing method and device

Abstract

The application provides a financial message pushing method and device, relates to the technical field of artificial intelligence, and can be applied to the technical field of finance or other technical fields. The financial message pushing method comprises the following steps: determining user reading content feature similarity data according to a pre-constructed financial message graph and user reading content data; constructing a financial relationship network according to the user reading content feature similarity data and user financial social strength data; determining a group to which a target user belongs according to the financial relationship network, and pushing financial messages of the group to the target user. The application can improve the accuracy of financial message pushing and achieve the effect of thousands of people with thousands of faces.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of artificial intelligence, in particular to a financial message pushing method and device. BACKGROUND

[0002] Many existing financial message pushing schemes are realized by recommendation algorithms and Apriori algorithms, which provide strong support for message recommendation prediction. However, the above financial message news pushing scheme is broad and does not consider the personal social circle and interpersonal relationship factors of the user. The social circle, interpersonal relationship and the received message of the user are closely related, and people tend to be interested in the message of their industry circle, and the reading hobby is also influenced by people around. The existing technology does not consider these interpersonal relationship factors. SUMMARY

[0003] The main purpose of the embodiment of the present application is to provide a financial message pushing method and device to improve the accuracy of financial message pushing and achieve the effect of thousands of people with thousands of faces.

[0004] In order to achieve the above purpose, the embodiment of the present application provides a financial message pushing method, comprising:

[0005] According to the pre-constructed financial message graph and the user reading content data, determine the user reading content feature similarity data;

[0006] According to the user reading content feature similarity data and the user financial social strength data, construct a financial relationship network;

[0007] According to the financial relationship network, determine the group to which the target user belongs, and push the financial message of the group to the target user.

[0008] In one embodiment, constructing a financial message graph comprises:

[0009] According to the financial data, constructing financial message atomic concept layer data, and classifying the financial message atomic concept layer data;

[0010] According to the financial data, constructing financial message general concept layer data;

[0011] According to the financial message general concept layer data, the classified financial message atomic concept layer data and the preset financial message classification tree layer data, constructing a financial message graph.

[0012] In one embodiment, the financial data includes structured financial data and unstructured financial data.

[0013] According to the financial data, constructing financial message atomic concept layer data comprises:

[0014] According to the unstructured financial data, financial data word vectors and entity recognition data are obtained;

[0015] According to the structured financial data, the financial data word vectors and the entity recognition data, financial message atomic concept layer data is constructed.

[0016] In one embodiment, classifying the financial message atomic concept layer data comprises:

[0017] According to a preset financial message classification tree layer data, the financial message atomic concept layer data is initially classified;

[0018] According to the unstructured financial data, syntax dependency analysis data is obtained to determine superior-inferior word pairs;

[0019] According to the superior-inferior word pairs, the financial message atomic concept layer data after the initial classification is reclassified.

[0020] In one embodiment, determining the user reading content feature similarity data according to the pre-constructed financial message graph and the user reading content data comprises:

[0021] According to the user reading content data, financial message vectors are constructed, and the financial message vectors are mapped to the financial message graph to obtain financial message path lengths;

[0022] According to the financial message path lengths, the user reading content feature similarity data is determined.

[0023] In one embodiment, it further comprises:

[0024] According to the user transaction data, user fund transaction intimacy is determined;

[0025] According to the user social data, user social behavior intimacy is determined;

[0026] According to the user published information data and the user searched data, user influence data is determined;

[0027] According to the user fund transaction intimacy, the user social behavior intimacy and the user influence data, user financial social strength data is determined.

[0028] The embodiment of the present application further provides a financial message pushing device, comprising:

[0029] A reading content feature similarity module is configured to determine user reading content feature similarity data according to a pre-constructed financial message graph and user reading content data;

[0030] A financial relationship network module is configured to construct a financial relationship network according to the user reading content feature similarity data and the user financial social strength data;

[0031] The financial message pushing module is configured to determine a group to which the target user belongs according to the financial relationship network, and push a financial message of the group to the target user.

[0032] In one embodiment, the system further comprises:

[0033] The classification module is configured to construct financial message atomic concept layer data according to the financial data, and classify the financial message atomic concept layer data.

[0034] The general concept layer data module is configured to construct financial message general concept layer data according to the financial data.

[0035] The financial message graph module is configured to construct a financial message graph according to the financial message general concept layer data, the classified financial message atomic concept layer data, and preset financial message classification tree layer data.

[0036] In one embodiment, the financial data comprises structured financial data and unstructured financial data.

[0037] The classification module comprises:

[0038] The unstructured financial data processing unit is configured to obtain financial data word vectors and entity recognition data according to the unstructured financial data.

[0039] The atomic concept layer data unit is configured to construct financial message atomic concept layer data according to the structured financial data, the financial data word vectors, and the entity recognition data.

[0040] In one embodiment, the classification module further comprises:

[0041] The primary classification unit is configured to perform primary classification on the financial message atomic concept layer data according to the preset financial message classification tree layer data.

[0042] The hyponym-hypernym pair unit is configured to obtain syntactic dependency analysis data according to the unstructured financial data to determine a hyponym-hypernym pair.

[0043] The secondary classification unit is configured to perform secondary classification on the financial message atomic concept layer data that has been subjected to the primary classification according to the hyponym-hypernym pair.

[0044] In one embodiment, the reading content feature similarity module comprises:

[0045] The path length unit is configured to construct a financial message vector according to the user reading content data, and map the financial message vector to the financial message graph to obtain a financial message path length.

[0046] The reading content feature similarity unit is configured to determine user reading content feature similarity data according to the financial message path length.

[0047] In one embodiment, the method further comprises:

[0048] a fund transaction closeness module configured to determine fund transaction closeness of the user according to the user transaction data;

[0049] a social behavior closeness module configured to determine social behavior closeness of the user according to the user social data;

[0050] a user influence data module configured to determine user influence data of the user according to the user published information data and the user searched data;

[0051] a financial social strength data module configured to determine user financial social strength data of the user according to the fund transaction closeness of the user, the social behavior closeness of the user and the user influence data.

[0052] The embodiment of the present application also provides a computer device, comprising a memory, a processor and a computer program stored in the memory and running on the processor, and the processor implements the steps of the financial message pushing method when executing the computer program.

[0053] The embodiment of the present application also provides a computer readable storage medium, which stores a computer program, and the computer program implements the steps of the financial message pushing method when executed by a processor.

[0054] The embodiment of the present application also provides a computer program product, comprising computer programs / instructions, and the computer programs / instructions implement the steps of the financial message pushing method when executed by a processor.

[0055] The financial message pushing method and device of the embodiment of the present application first determine user reading content feature similarity data according to a pre-constructed financial message graph and user reading content data, then construct a financial relationship network according to the user reading content feature similarity data and user financial social strength data, and finally determine a group to which a target user belongs according to the financial relationship network to push financial messages of the group to the target user, so that the accuracy of financial message pushing can be improved and the effect of different people having different experiences can be achieved. BRIEF DESCRIPTION OF DRAWINGS

[0056] In order to more clearly illustrate the technical solutions of the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed in the embodiment description. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0057] Figure 1 is a flowchart of the financial message pushing method in the embodiment of the present application;

[0058] Figure 2 is a flowchart of a financial message pushing method in another embodiment of the present application;

[0059] Figure 3 is a flowchart of constructing a financial message graph in an embodiment of the present application;

[0060] Figure 4 is a flowchart of constructing financial message atomic concept layer data in an embodiment of the present application.

[0061] Figure 5 is a flowchart of classifying financial message atomic concept layer data in an embodiment of the present application;

[0062] Figure 6 is a flowchart of S101 in an embodiment of the present application;

[0063] Figure 7 is a flowchart of determining user financial social strength data in an embodiment of the present application;

[0064] Figure 8 is a processing schematic diagram of financial data in an embodiment of the present application;

[0065] Figure 9 is a schematic diagram of a financial message graph in an embodiment of the present application;

[0066] Figure 10 is a schematic diagram of constructing financial message atomic concept layer data in another embodiment of the present application;

[0067] Figure 11 is a schematic diagram of classifying financial message atomic concept layer data in another embodiment of the present application;

[0068] Figure 12 is a schematic diagram of constructing financial message general concept layer data in an embodiment of the present application;

[0069] Figure 13 is a schematic diagram of constructing a financial relationship network graph based on interpersonal relationships in an embodiment of the present application;

[0070] Figure 14 is a structural block diagram of a financial message pushing device in an embodiment of the present application;

[0071] Figure 15 is a structural block diagram of a computer device in an embodiment of the present application. DETAILED DESCRIPTION

[0072] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0073] Those skilled in the art will recognize that embodiments of the present invention can be implemented as a system, apparatus, device, method, or computer program product. Therefore, this disclosure can be specifically implemented in the following forms: entirely hardware, entirely software (including firmware, resident software, microcode, etc.), or a combination of hardware and software.

[0074] The acquisition, storage, use, and processing of data in this application all comply with the relevant provisions of national laws and regulations.

[0075] Given that existing technologies do not consider the factors of interpersonal networks, resulting in overly broad financial message pushes, embodiments of the present invention provide a financial message push method and apparatus to improve the accuracy of financial message pushes and achieve a personalized effect. The present invention will be described in detail below with reference to the accompanying drawings.

[0076] Figure 1 This is a flowchart of the financial message push method in an embodiment of the present invention. Figure 2 This is a flowchart of a financial message push method according to another embodiment of the present invention. Figures 1-2 As shown, financial message push methods include:

[0077] S101: Determine the similarity data of user reading content features based on the pre-constructed financial message graph and user reading content data.

[0078] Figure 3 This is a flowchart illustrating the construction of a financial message graph in an embodiment of the present invention. For example... Figure 3 As shown, constructing a financial message graph includes:

[0079] S201: Construct the atomic concept layer data of financial messages based on financial data, and classify the atomic concept layer data of financial messages.

[0080] Financial data includes both structured and unstructured financial data.

[0081] Table 1

[0082]

[0083]

[0084]

[0085] Table 1 is a schematic table of structured financial data. As shown in Table 1, the structured financial data is derived from the user base information and user consumption information of enterprises that are already stored in a database, and the unstructured financial data includes:

[0086] 1. New information of finance in reports, books, articles and news is crawled by using a crawler (such as a regular expression of Python), such as financial news (including news title, news content, date of issue and URL link), research reports of securities companies, academic journals, column information, information of investment institutions (including information such as institution name, introduction, industry, scale and round), investment events (including event information, investment party, financing party, financing event, round and amount), company announcements and financial blogs.

[0087] 2. Financial public data sets that can be directly downloaded on the network, such as new words / popular words in the financial field (including financial news and World Bank development database) and specialized vocabulary in the financial field (such as financial encyclopedia entries and authoritative dictionaries).

[0088] 3. Enterprise business information is crawled by using a crawler.

[0089] 4. Behavior data generated by users in a news community is obtained by using a network crawler tool, as shown in Table 2.

[0090] Table 2

[0091]

[0092] Figure 8 is a schematic diagram of processing of financial data in an embodiment of the present application. As shown in Figure 8 , for some PDF category files and video and audio category financial information, text information can be extracted by using an OCR visual processing algorithm, and audio and video information can be extracted by using a speech recognition algorithm and converted into text information for subsequent processing operations.

[0093] Figure 4 is a flowchart of constructing financial message atomic concept layer data in an embodiment of the present application. Figure 10 is a schematic diagram of constructing financial message atomic concept layer data in another embodiment of the present application. As shown in Figure 4 and Figure 10 , constructing financial message atomic concept layer data according to financial data includes:

[0094] S301: Obtain financial data word vectors and entity recognition data according to unstructured financial data.

[0095] The application converts non-structured text data into structured data by using NLP technology. On this basis, the NLP structured data is converted into a spatial vector form, preparing for the next step of constructing a financial message graph based on a quantitative factor.

[0096] Considering that the analysis accuracy of complex long sentences is very low, the application first selects a comma as a punctuation symbol to divide large-scale unstructured financial data separated by one or more commas. Then, the divided data is subjected to word segmentation and named entity recognition (NER). The application uses a conditional random field (CRF) model and loads a corpus to perform word segmentation on the divided data to improve the accuracy of word segmentation.

[0097] After word segmentation, the data should be converted into a computable and structured financial data word vector, i.e., a matrix. The application uses a Word2Vec model to generate a word vector, which can be used to map each word and obtain a high-dimensional vector (word vector, Word Embedding) representation of each word by using a Word2Vec algorithm.

[0098] S302: Constructing financial message atomic concept layer data according to structured financial data, financial data word vectors, and entity recognition data.

[0099] In specific implementation, the structured financial data is first stored in a database to construct a preliminary atomic concept layer, and then financial message atomic concept layer data is constructed according to the financial data word vectors and the entity recognition data

[0100] Figure 5 is a flowchart of classifying financial message atomic concept layer data in an embodiment of the application. Figure 11 is a schematic diagram of classifying financial message atomic concept layer data in another embodiment of the application. As shown in Figure 5 and Figure 11 Classifying the financial message atomic concept layer data includes:

[0101] S401: Initially classifying the financial message atomic concept layer data according to preset financial message classification tree layer data.

[0102] First, define a first-level classification, in which the first-level content classification specially involved in the field of financial messages includes "macro factor", "industry factor", "fundamental factor", "market factor", "technical factor", "behavior factor", and "event factor". The hierarchical clustering (AGENS) method is used to cluster the atomic concept layer data, and then the clustered clusters are labeled with a first-level classification.

[0103] S402: Obtaining syntax dependency analysis data from unstructured financial data to determine upper and lower word pairs.

[0104] When the number of phrases under a primary classification is mined to a certain amount, the phrases need to be classified into different levels of categories, and each primary classification will be further classified into secondary and tertiary classifications. The present application uses a Bootstrapping algorithm to extract the hyponym-hypernym pairs from the text sentences.

[0105] In a specific implementation, the dependency relationship between the phrases in a sentence can be determined according to the syntactic dependency analysis data to define the relation extraction rules.

[0106] Table 3

[0107]

[0108]

[0109] Table 3 is a schematic table of the rule templates. As shown in Table 3, the present application extracts the entities constituting the hyponym-hypernym relationship by using the rule templates to discover different classification levels. The NP-chunks in the templates are noun phrase chunks, and the NP-chunks are generally smaller fragments than the complete noun phrases, and there is no overlap between the NP-chunks. Then, a hyponym-hypernym relation extraction seed set is defined, which is a set of entity relation patterns stored in a triple format (<entity, relation, entity>), and the seed set is used as the initial triple for the training of Bootstrapping. The algorithm detects the sentences of the corpus according to the predefined extraction rules, extracts the hyponym-hypernym pairs in the matching sentences, and processes the hyponym-hypernym pairs into the triple format of <entity, relation, entity> and stores them in the pattern library. The above process is repeatedly performed until no new hyponym-hypernym pairs are generated. Accordingly, the secondary and tertiary classifications and the like of the primary classification can be obtained.

[0110] S403: reclassifying the financial message atomic concept layer data that has been initially classified according to the hyponym-hypernym pairs.

[0111] S202: constructing the financial message general concept layer data according to the financial data.

[0112] Figure 12 FIG. 1 is a schematic diagram of constructing the financial message general concept layer data in an embodiment of the present application. As shown in FIG. 1, first, the AutoPhrase framework is used to automatically mine phrases from the unstructured financial data, and a list of phrases arranged in descending order of quality is output. Figure 12

[0113] Table 4

[0114] [Location] [Time] for [Technology Factor] [Industry Factor] for [Technology Factor]

[0115] Table 4 is a pattern schematic table in an embodiment of the present application. As shown in Table 4, the present application generates the financial message general concept layer data by combining the phrases according to Table 4.

[0116] ​S203: constructing a financial message graph according to the financial message general concept layer data, the classified financial message atomic concept layer data and the preset financial message classification tree layer data.

[0117] The existing knowledge graph storage takes <entity, relation, entity> as a basic expression mode, which is called a triple. The general knowledge graph mainly uses a triple structure, and lacks necessary depth and breadth. In order to better describe and understand the user's demand, solve the semantic gap generated by the algorithm to the user's demand in the subsequent mining process, the present application proposes a tree-style graph structure, which explicitly expresses the user's financial message acquisition demand as a node in the graph. Figure 9 is a schematic diagram of the financial message graph in the embodiment of the present application. As shown in Figure 9 , the financial message graph structure mainly consists of three parts, including an atomic concept layer, a general concept layer and a classification tree layer.

[0118] Atomic concept layer: store atomic concept data. In order to refine the user's financial message demand to the word granularity, use these fine-grained words to more systematically describe the financial message, and store these fine-grained words in the nodes of the atomic concept layer. For example, for the concept "A enterprise of the home appliance industry announced that B enterprise intends to acquire 5% of the company's shares for 42 billion yuan", it can be expressed as "industry factor: home appliance industry & enterprise: A enterprise & enterprise: B enterprise & event: share acquisition". Here, "home appliance industry", "A enterprise", "B enterprise" and "share acquisition" are all atomic words.

[0119] General concept layer: each node of this layer stores financial related phrases or concepts, such as "share acquisition" and "economic growth and consumption upgrading", etc. These phrases and names are not formally defined by dictionaries and encyclopedias, but are always mentioned by users. Many times, the user's demand for financial messages is a "scene" or "problem", not a search for financial terms that can help solve the problem, so the present application further generalizes the definition of the user's financial message demand to a concept.

[0120] Classification tree layer: In order to better manage the atomic concept layer, the present application constructs a classification tree system, which is a tree structure (more than 1 layer), and each classification node corresponds to the classification of the atomic concept layer node connected thereto. The system is not limited to the financial field, but is served for the financial message field. In this layer, the present application defines "time", "region", "industry factor", "market factor", "event factor" and "technology factor" as primary classifications, and further subdivides sub-classifications under each classification to form a tree structure, for example, the industry factor can be subdivided into the technology industry, real estate industry, home appliance industry, electronic information industry, chemical industry, automobile industry, financial industry, agriculture, forestry, animal husbandry, fishery industry, pharmaceutical industry, metallurgy industry, textile industry, machinery industry, paper industry packaging industry, building material industry, commercial industry and comprehensive industry. The above-mentioned "home appliance industry", "A enterprise", "B enterprise", "transferred shares" belong to "industry factor-home appliance industry", "company", "company" and "event factor-buying and selling" respectively. Meanwhile, different classifications have different relationships, for example, "place-country-China" and "technology factor-listing method" define a "applicable to" relationship, and the relationship is stored in the form of a triple, such as <listing method, applicable to, China>.

[0121] Figure 6 is a flowchart of S101 in an embodiment of the present application. As shown in Figure 6 , S101 includes:

[0122] S501: constructing a financial message vector according to the user reading content data, and mapping the financial message vector to the financial message graph to obtain the financial message path length.

[0123] In specific implementation, the user reading content data recently read by the user is obtained through a crawler, the words in each message are extracted and a message vector E={e1, e2, …, en} is constructed, wherein the message vector E is a message, and each feature e n is a key word obtained after the message is processed by word segmentation (the importance is measured according to TF-IDF). Each feature in {e1, e2, …, en} is corresponded to the atomic concept layer of the financial message graph and mapped to the category C={c1, c2, …, c n n} of the classification tree layer. If the user reading content data is highly similar to the elements of the existing general concept layer, then the words of the atomic concept layer connected by the elements of the general concept layer can be directly used; if the user reading content data cannot be found in the general concept layer, then the user reading content data needs to be added to the training set of the financial knowledge graph, and a new financial support graph is trained again, so that the new reading content of the user can be matched to the elements of the general concept layer. The link length of the feature e in the category C is the path distance from the root node of the classification tree layer to the position of the atomic concept layer node of the feature, that is:

[0124]

[0125] Wherein, Path(e i ,e j ) represents the i-th feature e i With the j-th feature e j The length of the financial message path between them For e i The path length from the root node of the category tree to the node in the atomic concept layer where the feature resides. For e j The path length from the root node of the category tree to the node in the atomic concept layer where the feature resides. i and e i The path distance to itself is 0.

[0126] S502: Determine the similarity data of user reading content features based on the path length of financial messages.

[0127] In practical implementation, e can be obtained through the following formula. i and e j Similarity between them:

[0128] Semantic_dist(e i ,e j ) = max{Path(e i ,e j ),1};

[0129] Among them, Semantic_dist(e i ,e j ) for e i and e j The similarity between them.

[0130] In the message dataset D = {E1, E2, ... E} m In the context of}, each message E is set. m The number of features is equal, meaning that the most important h features are selected for each message, and E m There is no overlap between the two messages. Therefore, we can define a two-dimensional matrix to represent the semantic similarity between the two messages: `Semantic_Matrix = [h*h]`. Each cell of the matrix represents the similarity between a feature of message E1 and all features of message E2, calculated using the `Semantic_dist` formula. Then, the resulting two-dimensional array is normalized. The specific steps are as follows:

[0131] 1. Get the maximum value of the column direction of a two-dimensional array.

[0132] 2. Obtain the lg value of each element of the two-dimensional array: lg(x+1); wherein x is an element in the two-dimensional array.

[0133] 3. Obtain the lg value of the maximum value in the column direction of the two-dimensional array: lg(x+1); wherein x is the maximum value in the column direction of the two-dimensional array. max max

[0134] 4. Logarithmic normalization is performed on the two-dimensional array using lg: x normalized =lg(x+1) / lg(x+1). max

[0135] Since each user reads more than one message, multiple two-dimensional matrices can be obtained. The multiple two-dimensional matrices are added, and then each element in the finally obtained two-dimensional array is added, and finally a natural number can be obtained, which is the user reading content feature similarity score, so as to obtain the reading content feature similarity between users.

[0136] The present application limits the time range to the last year, compares the position relationship of the current reading content feature similarity and the three-part position of the past year, that is, the fluctuation, and thus obtains the reading content feature similarity in the last year.

[0137] S102: Construct a financial relationship network according to the user reading content feature similarity data and the user financial social strength data.

[0138] Figure 13 is a schematic diagram of constructing a financial relationship network based on interpersonal relationship in an embodiment of the present application. As shown in the figure, in specific implementation, the interpersonal relationship between entities can be analyzed according to the user reading content feature similarity data and the user financial social strength data, the relationship edges between entities are given weights, and a financial relationship network based on interpersonal relationship is constructed. Figure 13

[0139] Figure 7 is a flowchart for determining user financial social strength data in an embodiment of the present application. As shown in the figure, the financial message pushing method further comprises: Figure 7

[0140] S601: Determine the user fund transaction closeness according to the user transaction data.

[0141] ​​​​​In specific implementation, the closeness can be divided into different levels, and a Likert 5-point scale is adopted, in which 1-5 represent very low closeness to very high closeness. The labels corresponding to 5, 4, 3, 2 and 1 are “very close”, “close”, “relatively close”, “general” and “not close” respectively; the relationship degree of the longest relationship chain in the group is not less than 2 degrees and not more than 4 degrees; the case of relationship closed loop is not considered.

[0142] If there is frequent transaction behavior between the transaction accounts held by two users, there may be a close relationship between them. Assuming that both transaction parties are normal settlement accounts, the user transaction data includes one-way transaction frequency, one-way transaction total number, one-way transaction total amount and transaction timeliness between the two parties. Among them, the transaction timeliness needs to consider two factors: “average value of adjacent transfer interval days” and “the interval days of the last adjacent transfer”. If “average value of adjacent transfer interval days” > “interval days of the last two adjacent transfers”, the user fund transaction closeness decreases; if “average value of adjacent transfer interval days” < “interval days of the last two adjacent transfers”, the user fund transaction closeness increases; if “average value of adjacent transfer interval days” = “interval days of the last adjacent transfer”, the user fund transaction closeness is unchanged.

[0143] S602: Determine the user social behavior closeness according to the user social data.

[0144] In specific implementation, the user social behavior closeness can be determined according to the degree of interaction between the users online and the frequency of offline meeting.

[0145] 1. The social behavior mediated by the social platform is weighted by the reading / collection / forwarding number between the two parties, and the total one-way reading / collection / forwarding number between the two parties needs to be counted respectively. For example, the interaction behavior is mutual, and the degree of interaction between user u and user v is defined as Interaction(u, v):

[0146]

[0147] Among them, Retransmit(u, v) is the number of times that user u forwards the post of user v, Retransmit(v, u) is the number of times that user v forwards the post of user u, Comment(u, v) is the number of times that user u comments on the post of user v, Comment(v, u) is the number of times that user v comments on the post of user u, Read(u, v) is the number of times that user u reads the post of user v, and Read(v, u) is the number of times that user v reads the post of user u.

[0148] 2. Offline social behavior is weighted by meeting frequency. Although this social behavior cannot be directly obtained from statistical data, it can be inferred from some electronic payment behavior, such as a payer purchasing more than one movie ticket, so the payer is likely to have a social activity with others to watch a movie. The payer can know who the payer interacts with from the level of viewer information when the payer purchases a movie ticket. Similarly, it can be extended to the scenes of scenic tickets and restaurant reservations, and these data can be obtained from various scenes in the digital bank ecosystem.

[0149] S603: Determine the user influence data according to the user publishing information data and the user being searched data.

[0150] First, the intuitive influence can be measured by the average number of times the subject is searched in the search engine, which is defined as BeAvgSearch(k) in the present application. In addition, the economic and financial topics are extracted from the news data using the LDA topic model (Latent Dirichlet Allocation), and then the number of positive comments, the number of forwards, the number of likes, and the number of collections obtained by the user using the forum post are counted. The more the number is, the greater the positive influence of the information published by the user on other users.

[0151] In specific implementation, the user positive influence can be determined by the following formula:

[0152]

[0153] wherein PositivePost(k) is the user positive influence of the kth user, Num positiveAvgComment(w) is the number of positive comments of the post w of the user, Num retransmits(w) is the number of forwards of the post w of the user, Num like(w) is the number of likes of the post w of the user, Num collect(w) is the number of collections of the post w of the user, and Total represents the total number of posts published by the user.

[0154] The user negative influence can be determined by the following formula:

[0155]

[0156] wherein NegativePost(k) is the user negative influence of the kth user, Num negativeAvgComment(w) is the number of negative comments of the post w of the user, Num unlike(w) is the number of dislikes of the post w of the user.

[0157] In summary, the user influence data is:

[0158] PostIncluencePower(k) = PositivePost(k) - NegativePost(k) + BeAvgSearch(k);

[0159] Wherein, PostIncluencePower(k) is the user influence data of the kth user, and BeAvgSearch(k) is the user search data of the kth user.

[0160] S604: Determine the user financial social strength data according to the user fund transaction intimacy, the user social behavior intimacy and the user influence data.

[0161] S103: Determine the group where the target user is located according to the financial relationship network, and push the financial message of the group to the target user.

[0162] In specific implementation, the method of hierarchical clustering (AGENS) is used to mine the similar group of reading financial messages. Through clustering, a plurality of similar groups of reading financial messages can be obtained, so that the target user can be clustered, and the user reading the top financial messages of the group where the target user is located can be pushed to the target user.

[0163] Figure 1 The execution subject of the financial message pushing method shown can be a computer. Figure 1 As shown in the flow, the financial message pushing method of the embodiment of the application first determines the user reading content feature similarity data according to the pre-constructed financial message graph and the user reading content data, then constructs the financial relationship network according to the user reading content feature similarity data and the user financial social strength data, and finally determines the group where the target user is located according to the financial relationship network to push the financial message of the group to the target user, so that the accuracy of the financial message pushing can be improved, and the effect of thousands of people with thousands of faces can be achieved.

[0164] In summary, the financial message pushing method provided by the embodiment of the application has the following beneficial effects:

[0165] (1) The multi-level tree structure financial message graph is designed, which can better describe and understand the user's demand for obtaining financial messages;

[0166] (2) The financial relationship network is constructed by combining the financial message graph and the calculation of the financial message social relationship;

[0167] (3) The clustering is used to achieve the user demand.

[0168] Based on the same inventive concept, the embodiment of the present application also provides a financial message pushing device. Since the device solves problems in the same principle as the financial message pushing method, the implementation of the device can be referred to the implementation of the method, and the repeated parts will not be described here.

[0169] Figure 14 is a structural block diagram of the financial message pushing device in the embodiment of the present application. As shown in Figure 14 , the financial message pushing device comprises:

[0170] a reading content feature similarity module, configured to determine user reading content feature similarity data according to a pre-constructed financial message graph and user reading content data;

[0171] a financial relationship network module, configured to construct a financial relationship network according to the user reading content feature similarity data and user financial social strength data;

[0172] a financial message pushing module, configured to determine a group in which a target user is located according to the financial relationship network, and push a financial message of the group to the target user.

[0173] In one embodiment, the device further comprises:

[0174] a classification module, configured to construct financial message atomic concept layer data according to financial data, and classify the financial message atomic concept layer data;

[0175] a general concept layer data module, configured to construct financial message general concept layer data according to the financial data;

[0176] a financial message graph module, configured to construct a financial message graph according to the financial message general concept layer data, the classified financial message atomic concept layer data and preset financial message classification tree layer data.

[0177] In one embodiment, the financial data comprises structured financial data and unstructured financial data.

[0178] The classification module comprises:

[0179] an unstructured financial data processing unit, configured to obtain financial data word vectors and entity recognition data according to the unstructured financial data;

[0180] an atomic concept layer data unit, configured to construct financial message atomic concept layer data according to the structured financial data, the financial data word vectors and the entity recognition data.

[0181] In one embodiment, the classification module further comprises:

[0182] a primary classification unit, configured to perform primary classification on the financial message atomic concept layer data according to the preset financial message classification tree layer data.

[0183] a superordinate-subordinate word pair unit configured to obtain syntactic dependency analysis data from the unstructured financial data to determine superordinate-subordinate word pairs;

[0184] a re-classification unit configured to re-classify the financial message atomic concept layer data that has been subjected to the primary classification according to the superordinate-subordinate word pairs.

[0185] In one embodiment, the reading content feature similarity module comprises:

[0186] a path length unit configured to construct a financial message vector from the user reading content data, map the financial message vector to the financial message graph to obtain financial message path length;

[0187] a reading content feature similarity unit configured to determine user reading content feature similarity data from the financial message path length.

[0188] In one embodiment, the financial message push device further comprises:

[0189] a fund transaction closeness module configured to determine user fund transaction closeness from the user transaction data;

[0190] a social behavior closeness module configured to determine user social behavior closeness from the user social data;

[0191] a user influence data module configured to determine user influence data from the user published information data and the user searched data;

[0192] a financial social strength data module configured to determine user financial social strength data from the user fund transaction closeness, the user social behavior closeness and the user influence data.

[0193] In summary, the financial message push device according to the embodiments of the present application first determines user reading content feature similarity data from a pre-constructed financial message graph and user reading content data, then constructs a financial relationship network from the user reading content feature similarity data and the user financial social strength data, and finally determines a group to which a target user belongs from the financial relationship network to push financial messages of the group to the target user, which can improve the accuracy of financial message push and achieve the effect of one person one face.

[0194] The embodiments of the present application further provide a specific implementation of a computer device capable of implementing all steps of the financial message push method in the above embodiments. Figure 15 is a structural block diagram of the computer device in the embodiments of the present application, referring to Figure 15 , the computer device specifically comprises the following contents:

[0195] A processor 1501 and a memory 1502.

[0196] The processor 1501 is configured to invoke a computer program in the memory 1502, and the processor executes the computer program to implement all steps in the financial message pushing method in the above embodiments, for example, the processor executes the computer program to implement the following steps:

[0197] determining user reading content feature similarity data according to the pre-constructed financial message graph and user reading content data;

[0198] constructing a financial relationship network according to the user reading content feature similarity data and user financial social strength data;

[0199] determining a group that the target user belongs to according to the financial relationship network, and pushing a financial message of the group to the target user.

[0200] In summary, the computer device in the embodiment of the present application first determines user reading content feature similarity data according to the pre-constructed financial message graph and user reading content data, then constructs a financial relationship network according to the user reading content feature similarity data and user financial social strength data, and finally determines a group that the target user belongs to according to the financial relationship network to push a financial message of the group to the target user, which can improve the accuracy of financial message pushing and achieve the effect of thousands of people with thousands of faces.

[0201] The embodiment of the present application also provides a computer readable storage medium capable of implementing all steps in the financial message pushing method in the above embodiments, and the computer readable storage medium stores a computer program, and the computer program is executed by the processor to implement all steps in the financial message pushing method in the above embodiments, for example, the processor executes the computer program to implement the following steps:

[0202] determining user reading content feature similarity data according to the pre-constructed financial message graph and user reading content data;

[0203] constructing a financial relationship network according to the user reading content feature similarity data and user financial social strength data;

[0204] determining a group that the target user belongs to according to the financial relationship network, and pushing a financial message of the group to the target user.

[0205] In summary, the computer readable storage medium of the embodiment of the present application first determines the user reading content feature similarity data according to the pre-constructed financial message graph and the user reading content data, then constructs the financial relationship network according to the user reading content feature similarity data and the user financial social strength data, and finally determines the group where the target user is located according to the financial relationship network to push the financial message of the group to the target user, which can improve the accuracy of the financial message pushing and achieve the effect of thousands of people with thousands of faces.

[0206] The embodiment of the present application also provides a computer program product capable of implementing all steps in the financial message pushing method in the above-mentioned embodiment, and the computer program product comprises computer programs / instructions which, when executed by a processor, implement all steps of the financial message pushing method in the above-mentioned embodiment, for example, the processor implements the following steps when executing the computer program:

[0207] determining the user reading content feature similarity data according to the pre-constructed financial message graph and the user reading content data;

[0208] constructing the financial relationship network according to the user reading content feature similarity data and the user financial social strength data;

[0209] determining the group where the target user is located according to the financial relationship network, and pushing the financial message of the group to the target user.

[0210] In summary, the computer program product of the embodiment of the present application first determines the user reading content feature similarity data according to the pre-constructed financial message graph and the user reading content data, then constructs the financial relationship network according to the user reading content feature similarity data and the user financial social strength data, and finally determines the group where the target user is located according to the financial relationship network to push the financial message of the group to the target user, which can improve the accuracy of the financial message pushing and achieve the effect of thousands of people with thousands of faces.

[0211] The above-described specific embodiments further illustrate the purpose, technical solutions and beneficial effects of the present application, and it should be understood that the above-described embodiments are only specific embodiments of the present application and are not used to limit the protection scope of 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.

[0212] Those skilled in the art will also understand that the various illustrative logical blocks, units, and steps listed in the embodiments of the present invention can be implemented by electronic hardware, computer software, or a combination of both. To clearly demonstrate the interchangeability of hardware and software, the functions of the various illustrative components, units, and steps described above have been generally described. Whether such functionality is implemented through hardware or software depends on the specific application and the overall system design requirements. Those skilled in the art can implement the described functions using various methods for each specific application, but such implementation should not be construed as exceeding the scope of protection of the embodiments of the present invention.

[0213] The various illustrative logic blocks, units, or devices described in the embodiments of this invention can be implemented or operate the described functions using a general-purpose processor, digital signal processor, application-specific integrated circuit (ASIC), field-programmable gate array or other programmable logic device, discrete gate or transistor logic, discrete hardware components, or any combination thereof. The general-purpose processor can be a microprocessor; alternatively, it can be any conventional processor, controller, microcontroller, or state machine. The processor can also be implemented using a combination of computing devices, such as a digital signal processor and a microprocessor, multiple microprocessors, one or more microprocessors combined with a digital signal processor core, or any other similar configuration.

[0214] The steps of the methods or algorithms described in the embodiments of this invention can be directly embedded in hardware, a software module executed by a processor, or a combination of both. The software module can be stored in RAM, flash memory, ROM, EPROM, EEPROM, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium in the art. Exemplarily, the storage medium can be connected to the processor so that the processor can read information from and write information to the storage medium. Optionally, the storage medium can also be integrated into the processor. The processor and storage medium can be housed in an ASIC, which can be housed in a user terminal. Optionally, the processor and storage medium can also be housed in different components of the user terminal.

[0215] In one or more exemplary designs, the functions described can be implemented in hardware, software, firmware, or any combination thereof. If implemented in software, the functions can be stored on or transmitted over as one or more instructions or code on a computer-readable medium. Computer-readable media includes both computer storage media and communication media including any medium that facilitates transfer of a computer program from one place to another. Storage media can be any available media that can be accessed by a computer. By way of example, such computer-readable media can comprise RAM, ROM, EEPROM, CD-ROM or other optical disk storage, magnetic disk storage or other magnetic storage devices, or any other medium that can be used to carry or store desired program code in the form of instructions or data structures and that can be accessed by a general-purpose or special-purpose computer, or a general-purpose or special-purpose processor. Also, any connection is properly termed a computer-readable medium. For example, if the software is transmitted from a website, server, or other remote source using a coaxial cable, fiber optic cable, twisted pair, digital subscriber line (DSL), or wireless technologies such as infrared, radio, and microwave, then the coaxial cable, fiber optic cable, twisted pair, DSL, or wireless technologies such as infrared, radio, and microwave are included in the definition of computer-readable medium. Disk and disc, as used herein, includes compact disc (CD), laser disc, optical disc, DVD, floppy disk, and Blu-ray disc where disks usually reproduce data magnetically, while discs reproduce data optically with lasers. Combinations of the above should also be included within the scope of computer-readable media.

Claims

1. A method for pushing financial messages, characterized in that, include: Financial message vectors are constructed based on user reading content data, and the financial message path length is obtained by mapping the financial message vectors to the financial message graph. The similarity data of user reading content features is determined based on the length of the financial message path. The construction of a financial information graph includes: A financial message atomic concept layer data is constructed based on financial data, and the financial message atomic concept layer data is classified; the financial data includes structured financial data and unstructured financial data; The data used to construct the atomic concept layer of financial messages based on financial data includes: Financial data word vectors and entity recognition data are obtained from the unstructured financial data; financial message atomic concept layer data are constructed based on the structured financial data, the financial data word vectors, and the entity recognition data; Construct a general concept layer of financial messages based on the aforementioned financial data; A financial message graph is constructed based on the general concept layer data of financial messages, the classified atomic concept layer data of financial messages, and the preset financial message classification tree layer data. A financial relationship network is constructed based on the similarity data of user reading content features and the user's financial social strength data. The data used to determine the strength of a user's financial social connections includes: Determine the user's financial transaction intimacy based on user transaction data; Determine the intimacy of users' social behaviors based on user social data; User influence data is determined based on user-posted information data and user search data. The user's financial social intensity data is determined based on the user's financial transaction intimacy, the user's social behavior intimacy, and the user's influence data. Based on the financial relationship network, the target user's group is determined, and financial messages from that group are pushed to the target user.

2. The financial message push method according to claim 1, characterized in that, The classification of the atomic concept layer data of the financial messages includes: The financial message atomic concept layer data is initially classified according to the preset financial message classification tree layer data; Syntactic dependency analysis data is obtained from the unstructured financial data to determine hyponym / hypernym pairs; The data at the atomic concept level of financial messages, which had been initially classified, were reclassified based on the hyponyms and hypernyms.

3. A financial message push device, characterized in that, include: The path length unit is used to construct a financial message vector based on the user's reading content data, and to map the financial message vector to the financial message graph to obtain the financial message path length. The reading content feature similarity unit is used to determine the similarity data of user reading content features based on the path length of financial messages; The classification module is used to construct the atomic concept layer data of financial messages based on financial data, and to classify the atomic concept layer data of financial messages; the financial data includes structured financial data and unstructured financial data; A general concept layer data module is used to construct general concept layer data for financial messages based on the financial data. The financial message graph module is used to construct a financial message graph based on the general concept layer data of financial messages, the classified atomic concept layer data of financial messages, and the preset financial message classification tree layer data. The classification module includes: An unstructured financial data processing unit is used to obtain financial data word vectors and entity recognition data based on the unstructured financial data; Atomic concept layer data unit, used to construct financial message atomic concept layer data based on the structured financial data, the financial data word vectors, and the entity recognition data; The financial relationship network module is used to construct a financial relationship network based on the similarity data of user reading content features and the user's financial social strength data. The Funds Transaction Intimacy module is used to determine the user's funds transaction intimacy based on the user's transaction data; The social behavior intimacy module is used to determine the intimacy of a user's social behavior based on the user's social data; The User Influence Data Module is used to determine user influence data based on user-posted information data and user search data. The Financial Social Intensity Data Module is used to determine the financial social intensity data of users based on the intimacy of user fund transactions, the intimacy of user social behavior, and user influence data. The financial message push module is used to determine the group to which the target user belongs based on the financial relationship network, and push financial messages of that group to the target user.

4. A computer device, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the financial message push method according to any one of claims 1 to 2.

5. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the financial message push method according to any one of claims 1 to 2.

6. A computer program product comprising a computer program / instructions, characterized in that, When the computer program / instructions are executed by the processor, they implement the steps of the financial message push method according to any one of claims 1 to 2.

Images