Search

CN-121998759-A - Transaction opponent recommendation method and system based on multi-mode data

CN121998759ACN 121998759 ACN121998759 ACN 121998759ACN-121998759-A

Abstract

The invention discloses a transaction opponent recommending method and a transaction opponent recommending system based on multi-mode data, and relates to the technical field of financial software development. The method comprises the steps of obtaining multi-mode data, carrying out data fusion on the multi-mode data, configuring a multi-dimensional portrait tag for each candidate trading opponent according to the fusion data, comprising an accurate tag and a fuzzy tag, constructing a trading opponent matrix comprising tag information of all candidate trading opponents according to the multi-dimensional portrait tag, obtaining trading feature elements of a bond product to be traded after analyzing when receiving an investment instruction, and outputting target trading opponent information and/or sending a price inquiring instruction to the target trading opponent after determining the target trading opponent matched with the bond product according to the multi-dimensional matching model when acquiring price inquiring requirements of a trading clerk for the bond product. The invention can obviously improve the efficiency and success rate of bond price inquiring transaction.

Inventors

  • WU WEI
  • ZHONG XIANGYAN
  • LOU KANGHUA
  • WANG ZHIHAN
  • ZHOU YOU

Assignees

  • 上海中汇亿达金融信息技术有限公司

Dates

Publication Date
20260508
Application Date
20251225

Claims (11)

  1. 1. A transaction opponent recommending method based on multi-mode data is characterized by comprising the following steps: acquiring multi-modal data, wherein the multi-modal data is multi-source data which is acquired from different channels and is related to bond transactions; the multi-dimensional portrait label comprises an accurate label and a fuzzy label, wherein the accurate label is used for identifying bond codes of expected transactions of candidate transaction opponents, and the fuzzy label is used for identifying bond preference characteristics of expected transactions of the candidate transaction opponents; Constructing a transaction opponent matrix comprising label information of all the candidate transaction opponents according to the multidimensional portrait labels of the candidate transaction opponents; When an investment instruction is received, analyzing the investment instruction to obtain transaction characteristic elements of a bond product to be transacted, wherein the transaction characteristic elements at least comprise bond codes; When acquiring the price inquiring requirement of the trader for the bond product, determining a target trade opponent matched with the bond product through a multidimensional matching model according to the trade opponent matrix, outputting target trade opponent information and/or sending a price inquiring instruction to the target trade opponent.
  2. 2. The method of claim 1, wherein the candidate counterparty is a friend of the trader in an address book of the associated chat tool, and wherein the constructed counterparty matrix is a friend matrix.
  3. 3. The method of claim 1, wherein the fuzzy tags include preference tags for identifying bond type information preferred by the candidate transaction opponents at the time of the transaction and behavior feature tags for identifying bond class information preferred by the candidate transaction opponents at the time of the transaction; Wherein the bond types are distinguished by a child level type of the issuing body of the bond, and the bond subclass is distinguished by a parent level type of the issuing body of the bond, the parent level type having one or more child level types thereunder.
  4. 4. The method of claim 3, wherein the multimodal data includes structured data and unstructured data; The structured data comprises historical transaction data and coupon data which adopt a preset standard data format; the unstructured data includes unstructured interactive data including chat data in an associated chat tool and unstructured real-time market data including intent offer data in the associated chat tool.
  5. 5. The method of claim 4, wherein configuring the multi-dimensional portrayal labels for each candidate transaction partner based on the fused data comprises: For each candidate transaction opponent, integrating real-time chat data and quotation data of the candidate transaction opponent in an associated chat tool, then carrying out data semantic analysis, judging whether the candidate transaction opponent has a bond code of expected transaction according to a semantic analysis result, judging that sometimes, taking the bond code as an accurate label of the candidate transaction opponent, and recording the data occurrence date; Integrating historical transaction data and bond data of the candidate transaction opponent, then carrying out data semantic analysis, judging whether the candidate transaction opponent has a bond type preferred according to semantic analysis results, judging that sometimes, taking the bond type as a preference label of the candidate transaction opponent, and recording data occurrence date; And integrating historical transaction data and quotation data of the candidate transaction opponent, then carrying out data semantic analysis, judging whether the candidate transaction opponent has a preferred bond class according to a semantic analysis result, judging that sometimes, taking the bond class as a behavior feature label of the candidate transaction opponent, and recording the data occurrence date.
  6. 6. The method of claim 5, wherein after configuring the multi-dimensional portrayal labels of each candidate transaction partner, constructing the transaction partner matrix based on the names and/or numbers of the candidate transaction partners, the multi-dimensional portrayal labels, the date of data occurrence, and the data type; The elements of the transaction opponent matrix at least comprise names and/or numbers of candidate transaction opponents, precision labels, preference labels, behavior characteristic labels, data occurrence dates and data type information.
  7. 7. The method of any of claims 1-6, wherein the matching of the candidate transaction opponent to the bond product is obtained by calculating a matching score, the multi-dimensional matching model being configured to calculate the matching score of the bond product to the candidate transaction opponent in a hierarchical matching manner combining exact matching and association rule matching; at this time, firstly, the accurate matching is performed between the bond code of the bond product and the accurate tag to obtain the matched trade opponents, and when the number of the exactly matched trade opponents does not meet the preset number requirement, the association rule matching of the tag is triggered to obtain more trade opponents matched with the bond product.
  8. 8. The method of claim 7, wherein the step of calculating a matching degree score of the bond product and the candidate opponent by means of hierarchical matching is as follows: S510, for each bond product in the investment instruction, inquiring in the trade opponent matrix based on the bond code of the bond product, acquiring accurate labels matched with the bond code, and extracting candidate trade opponent information to which the accurate labels belong to form trade opponent information to be matched; for each trade opponent to be matched, adjusting the weight of the accurate tag data of the trade opponent to be matched according to the tag factor parameter related to time, and calculating the matching degree score of the trade opponent to be matched and the bond product through a weighting algorithm according to the adjusted weight value; S520, judging whether the number of the traders to be matched reaches a preset number threshold N, wherein N is an integer greater than or equal to 2, executing step S530 when judging that the number reaches the preset number threshold N, and executing step S540 when judging that the number does not reach the preset number threshold N; s530, recommending and outputting the sorted first N trade opponents to be matched as target trade opponents, and ending; and S540, triggering association rule matching, at this time, after a transaction opponent-bond data set is constructed according to the accurate label and the fuzzy label data in the transaction opponent matrix, acquiring an associated bond product which has highest association with the bond product and appears in the accurate label data through a preset confidence algorithm, performing accurate matching through the bond code of the associated bond product to obtain a target transaction opponent related to the associated bond product, and recommending and outputting the target transaction opponent of the associated bond product as the target transaction opponent of the bond product.
  9. 9. The method of claim 8, wherein the time-dependent tagging factor comprises a time enhancement factor and a time decay factor, wherein the weight value corresponding to the time enhancement factor increases over time within a predetermined time range, and wherein the weight value corresponding to the time decay factor decreases over time within the predetermined time range; and configuring each data type as an item according to the data type information in the transaction opponent matrix, configuring the highest item score max_score for each data type item, and calculating the matching degree score of the transaction opponent to be matched and the bond product through a weighting algorithm, wherein the matching degree score comprises the following steps of: For each data type item, a single tag score lable _score for each precision tag is calculated as follows: After determining a weight value corresponding to a label factor of an accurate label according to a data occurrence date, multiplying the weight value by a label initial score preset by the accurate label to obtain a single label score lable _score of the accurate label, namely lable _score=label factor weight value; Adding tag scores of a plurality of accurate tags belonging to the same data type item to obtain each item score item_score, wherein the value of the item_score does not exceed the highest score max_score of the data type item configuration item, namely item_score=Min # ,max_score); Adding the scores of the plurality of data type items related to the bond product to obtain a bond total score bond score, the bond score having a value not exceeding a pre-configured score threshold MAX, i.e., bond score = Min # ,MAX)。
  10. 10. The method of claim 7, wherein the confidence algorithm is FP-Growth algorithm, and wherein the step of obtaining the associated bond product that has the highest association with the aforementioned bond product and that appears in the accurate label data is as follows: Constructing a trade opponent-bond data set according to the accurate label and the fuzzy label data in the trade opponent matrix, wherein the trade opponent-bond data set records historical bond data purchased by the candidate trade opponent and the candidate trade opponent; Calculating the support degree of each bond according to the frequency of occurrence of bonds in the transaction opponent-bond data set; Sorting the remaining bonds after filtering according to the support degree to obtain a sorted data set; Constructing a frequent item tree structure FP tree based on the ordered data set; Obtaining conditional pattern base data of the bond products in the investment instructions according to the built FP treee; Calculating the confidence of the related bonds according to the condition mode base data of the bond products and the support degree table of each bond; The method comprises the steps of sorting bonds according to the confidence coefficient of the bonds, taking the bonds with the largest confidence coefficient and appearing in the accurate label data as associated bond products, and accurately matching the associated bond products to obtain target transaction opponent information.
  11. 11. A transaction opponent recommendation system of an electronic transaction platform, comprising: The data acquisition module is used for acquiring multi-modal data, wherein the multi-modal data is multi-source data which is acquired from different channels and is related to bond transaction; The system comprises a tag configuration module, a multi-dimensional portrait tag and a multi-dimensional portrait tag, wherein the tag configuration module is used for carrying out data fusion on acquired multi-modal data and configuring a multi-dimensional portrait tag for each candidate transaction opponent according to the fusion data, the multi-dimensional portrait tag comprises an accurate tag and a fuzzy tag, the accurate tag is used for identifying bond codes of expected transactions of the candidate transaction opponent, and the fuzzy tag is used for identifying bond preference characteristics of expected transactions of the candidate transaction opponent; The matrix construction module is used for constructing a transaction opponent matrix comprising label information of all the candidate transaction opponents according to the multi-dimensional portrait labels of the candidate transaction opponents; the investment instruction analysis module is used for acquiring transaction characteristic elements of the bond products to be transacted after analyzing the investment instruction when receiving the investment instruction, wherein the transaction characteristic elements at least comprise bond codes; And the trade opponent matching module is used for outputting a target trade opponent and/or sending a price inquiring instruction to the target trade opponent after determining the target trade opponent matched with the bond product through the multi-dimensional matching model according to the trade opponent matrix when collecting the price inquiring requirement initiated by the trade person aiming at the bond product.

Description

Transaction opponent recommendation method and system based on multi-mode data Technical Field The invention relates to the technical field of financial software development, in particular to a transaction opponent recommendation method and system based on multi-mode data. Background With the networking and globalization trend of the financial industry, a transaction management system for managing data of market transactions among banks, such as ComStar system, has been developed for rapid change of market and business development of banks. The ComStar system covers all business varieties of the foreign currency trading platform of the Chinese foreign exchange trading center, the current coupon and the purchase return business of Shanghai and Shenzhen securities, the stock and derivative trading, the transfer and purchase return business of the ticket exchange, the national liability futures business of the Zhongjin and various off-line businesses, realizes the foreign currency integration and the front stage, middle stage and back stage straight-through processing, greatly reduces manual work and reduces operation risks. Meanwhile, the ComStar system can be seamlessly integrated with a foreign exchange transaction center transaction platform and a post-transaction platform, so that the through processing of transaction strategies, pre-approval, real-time credit control and transaction confirmation and fund clearing is realized. The inter-bank market and the exchange market are two major places for financial instrument transactions such as bonds, foreign exchange and the like. Currently, in some bond transactions (such as credit and debt transactions) in the inter-bank market and the exchange market, the transaction is mainly performed in a mode of double-sided negotiation, price inquiry and quotation, and the like, a trader inquires price to an opponent through a chat tool, and the selection of the trade opponent determines the efficiency and cost for achieving the exchange. This process is highly dependent on the personal experience and vein resources of the trader, with the following drawbacks: 1) Inefficiency-the trader needs to manually screen and memorize a large number of counter-party preferences, time consuming and laborious. 2) The subjectivity is strong, the matching process depends on personal experience, and high-quality opponents are easily missed or non-optimal opponents are selected. 3) The resources can not be shared, namely the resources in each trader can not be shared, new traders lack accumulation, work is difficult to be rapidly and effectively carried out, and the resources are easily taken away by personnel away from the staff. Aiming at the defects, the prior art provides some electronic transaction platforms capable of intelligently recommending transaction opponents, and more commonly, the related institution (transaction opponents) information is recommended to the transaction opponents after data statistics analysis is carried out by adopting some traditional data analysis tools or simple rule engines based on historical transaction data. However, the above-mentioned electronic transaction platform provides a list of institutions, which can only simply list related institutions (transaction opponents), lacks intelligent ordering and recommending functions, and still has an optimization space in terms of intellectualization and precision of the transaction opponents. Meanwhile, the scheme relies on historical data to perform static analysis, lacks dynamic processing capability on the data, cannot flexibly cope with dynamic changes of a transaction market, and therefore recommendation results possibly lag behind market reality. On the other hand, with the development and popularization of information processing means such as artificial intelligence, big data analysis, machine learning and the like, the prior art provides various technical schemes capable of carrying out automation, individuation and precision marketing decision and execution based on information such as user behavior data, transaction data, external environment variables and the like, the core content of the method generally comprises user portrait construction, behavior prediction modeling, marketing strategy generation, marketing content pushing and channel optimization and the like, and the method emphasizes that the method carries out fine identification and classification on users to carry out differential marketing by the technical means, mainly relates to data acquisition and processing, model construction and training, strategy decision generation, information pushing execution and effect tracking and feedback analysis technology, and is an important component closely combined with market popularization in digital enterprise operation. Based on the above technology, the prior art also provides some intelligent marketing decision schemes suitable for the financial field. For example, chinese patent ZL202510