TWM645430U - Fund Search Optimization System - Google Patents

Abstract

A fund search optimization system includes a storage device and a server. The server includes a processing unit and a first machine learning unit. The storage device stores fund information and fund name keyword data; when the processing unit determines that there is a corresponding search in the fund name keyword data When the fund name or keyword of the string is found, the processing unit generates search results based on the fund data corresponding to the fund name or keyword; when the processing unit determines that the fund name keyword data does not contain the fund name or keyword of the corresponding search string. , the first machine learning unit generates a similar search result based on at least one fund information corresponding to a fund name or keyword that is similar to the search string.

Description

Fund search optimization system

An optimization system, especially a fund search optimization system.

With the changes in social patterns, the investment trend has gradually become more popular, and people have begun to pay attention to financial planning in order to protect the quality of life after retirement. Due to the development and popularity of Internet technology, financial institutions have launched their own fund platforms, allowing people to use Searching and purchasing funds on your own online makes it more convenient to obtain fund information than relying on the introduction and promotion of investment consultants in the past.

However, although each fund product has a different International Securities Identification Code (ISIN Code), various financial institutions often modify the fund name or fund name of each fund product based on different considerations such as sales strategies, fund classifications, and platform management. The plan name and warning are additionally added, so that the same fund product often has different fund names on different platforms. People are accustomed to search by fund name when looking for fund products, but the differences in fund names between different platforms can easily cause people to be unable to search. to the corresponding fund products, thereby affecting the promotion and sales of fund products.

Therefore, in order to accurately push fund products for customers to purchase, how to improve the accuracy of fund searches is one of the problems that financial institutions currently need to solve.

In view of this, this new type of fund search optimization system provides a fund search optimization system that uses data to recognize the search string entered by the user and pushes relevant fund products to the user in order to improve the accuracy of fund search.

In order to achieve the aforementioned purpose, this new fund search optimization system receives a search string transmitted by a client device. The fund search optimization system includes: A storage device that stores a plurality of fund information and a fund name keyword data. The fund name keyword data includes the fund name of each fund product and at least one corresponding keyword; A server electrically connected to the storage device, including: A processing unit, electrically connected to the storage device, the processing unit determines whether there is at least one fund name or at least one keyword corresponding to the search string in the fund name keyword data. When the processing unit determines that the fund name keyword data When the at least one fund name or the at least one keyword corresponding to the search string is included, a search result is generated based on the at least one fund information corresponding to the at least one fund name or the at least one keyword, and the search result is Results are presented via a search page; and A first machine learning unit is electrically connected to the storage device and the processing unit. When the processing unit determines that the fund name keyword information does not include the at least one fund name or the at least one keyword corresponding to the search string, The processing unit provides the search string to the first machine learning unit. The first machine learning unit analyzes the fund name keyword information and at least one fund name or at least one keyword, and based on the search string similar to the At least one fund name or at least one fund information corresponding to the at least one keyword generates a similar search result, and the first machine learning unit transmits the similar search result to the processing unit, and the processing unit transmits the similar search result Presented through this search page.

The new fund search optimization system first compares the search string entered when the client device performs a fund search with the fund name keyword data through the processing unit to determine whether there is at least one fund product corresponding to the search string. And generate a search result based on the at least one fund product, and when the search string input by the client device cannot directly correspond to the fund name or keyword in the fund name keyword data due to differences in pinyin or word selection , the first machine learning unit further performs a similarity comparison between the search string and the fund name keyword data, and then generates a similar search result based on at least one fund product similar to the search string, which is similar to the conventional technology. Compared with this, in addition to presenting accurate search results to customers, this new model also presents similar search results to customers when the search string cannot directly correspond to the keyword information of the fund name, which helps to improve the benchmark of fund searches. , and increase the chance of users searching for intended products.

Please refer to Figure 1. The new fund search optimization system includes a storage device 10 and a server 20. The server 20 is electrically connected to the storage device 10. The fund search optimization system and a client device 30 communicate through a network. Through the communication connection, the client device 30 transmits a search string to the server 20 according to the user operation.

The storage device 10 stores a plurality of fund information, a weighted information and a fund name keyword information. Each fund information includes the fund name of a fund product, the International Securities Identification Code (ISIN Code), new companies, product content, subscription records and other information; the weighted data includes a plurality of weighted columns, each weighted column has a weight; the fund name keyword data includes each fund name and at least one corresponding keyword, and the same keyword can correspond to Multiple fund names, wherein the storage device 10 can be a random access memory (RAM), a read-only memory (ROM), a traditional hard disk (Hard Disk Drive, HDD), a solid-state disk (Solid-State Disk, SSD) and other electronic devices with data storage functions.

The server 20 is electrically connected to the storage device 10. The server 20 includes a processing unit 21, a first machine learning unit 22, a crawler unit 23 and a second machine learning unit 24. The server 20 receives data via a network. The search string transmitted by the client device 30 .

Referring further to FIG. 2 , the processing unit 21 is electrically connected to the storage device 10 . The processing unit 21 is provided with a search page 211 for the client device 30 to connect to. When the client device 30 executes the fund in the search page 211 When searching, the client device 30 enters the search string in a search field 212 on the search page 211 and transmits the search string to the processing unit 21 for reception. When the client device 30 has not yet executed a fund search, , which means that when the client device 30 has not input the search string, the processing unit 21 sorts the plurality of fund data according to the weighted data to generate a default display result, and presents the default display result on the search page. 211 in.

The processing unit 21 can update and manage the weighted data and the fund name keyword data, and after receiving the search string transmitted by the client device 30, the processing unit 21 determines whether there is a pair of tags in the fund name keyword data. At least one fund name or at least one keyword of the search string should be searched. When the processing unit 21 determines that the fund name keyword data contains the at least one fund name or the at least one keyword corresponding to the search string, the The processing unit 21 retrieves at least one fund information corresponding to the at least one fund name or at least one keyword from the storage device 10, and then sorts the at least one fund information according to the weighted information to generate a search result, and sorts the search result. The results are presented through the search page 211. When the processing unit 21 determines that the fund name keyword information does not contain any fund name or keyword corresponding to the search string, the processing unit 21 transmits the search string. to the first machine learning unit 22 for further similarity comparison.

The first machine learning unit 22 is electrically connected to the storage device 10 and the processing unit 21, and receives the search string from the processing unit 21, and the first machine learning unit 22 retrieves string features in the search string. , and through natural language processing technologies such as Levenshtein distance and Jaccard similarity coefficient, the string characteristics of the search string and the fund name keyword data are compared and analyzed for similarity, and the identification At least one fund name or at least one keyword that is similar to the search string, and at least one fund information corresponding to the at least one fund name or the at least one keyword is retrieved from the storage device 10, and then the at least one fund name is retrieved from the storage device 10. The data is sorted according to the weighted data to generate a similar search result, and the similar search result is transmitted to the processing unit 21 , and the processing unit 21 presents the similar search result through the search page 211 .

Among them, the first machine learning unit 22 uses natural language processing technology to perform similar comparison of the search string. For example, the first machine learning unit 22 can use homophones, approximate pronunciations, and phonetic symbols of the search string. Conduct semantic analysis on string features such as pinyin, numerical sorting, and English alphabetical sorting, and conduct a similar comparison between the string features of the search string and the fund name keyword data. When it is determined that there is at least one character in the fund name keyword data, When the similarity between the fund name or at least one keyword and the search string is greater than a preset value, the similar search results corresponding to the at least one fund name or the at least one keyword are generated, and the first machine learning unit 22 After the similar search results are generated, a new keyword is further created using the search string. The new keyword corresponds to the at least one fund product similar to the search string, and the fund name is updated based on the new keyword. keyword data, and then transmit the new keyword data of the fund name to the storage device 10 for storage.

On the other hand, when the first machine learning unit 22 determines that at least one fund name does not exist in the fund name keyword data or the similarity between at least one keyword and the search string is greater than a preset value, it represents the first The machine learning unit 22 determines that there is no data related to the search string in the fund name keyword data and cannot generate search results. Therefore, the first machine learning unit 22 further generates search exception data based on the search string, and The search anomaly data is stored in the storage device, where the search anomaly data can assist the administrator to understand the search anomaly situation and the corresponding search string, allowing the administrator to update and manage the fund name keywords based on the search anomaly data. material.

The crawler unit 23 is electrically connected to the storage device 10. The crawler unit 23 searches the fund names and international securities identification codes of various fund products in other fund websites through the Internet, and generates a crawler data. The crawler unit 23 converts the crawler The data is transmitted to the second machine learning unit 24, where the crawler unit 23 uses Python's Scrapy framework to crawl, and extracts information from other fund websites through Cascading Style Sheets (CSS) or XML Path (XML Path). data.

The second machine learning unit 24 is electrically connected to the storage device 10 and the crawler unit 23, and receives the crawler data transmitted by the crawler unit 23. The second machine learning unit 24 compares the crawler data with the fund name keyword data. Similarity comparison: Compare the fund names of other websites in the crawler data with the fund names or keywords in the fund name keyword data. When judging whether a fund name on other websites in the crawler data is similar to the fund name or keywords, When each fund name or each keyword in the fund name keyword data is similar, the second machine learning unit 24 sets a new keyword based on the fund name in the crawler data, and updates the fund based on the new keyword. name keyword data, and then transmits the new fund name keyword data to the storage device 10 for storage. The second machine learning unit 24 performs similarity comparison through Python's Natural Language Toolkit (NLTK). right.

Among them, the second machine learning unit 24 can identify fund products on other fund websites that have the same international securities identification code as the plurality of fund information through the international securities identification code. The second machine learning unit 24 can identify other fund products in the crawler data that have the same international securities identification code as the fund product based on the international security identification code of one of the fund products in the fund data, and based on the international security identification code of the fund product, For fund names of other fund products with the same international securities identification code, add the keywords corresponding to the fund product in the fund name keyword data.

Further, when the second machine learning unit 24 performs similar comparison, the second machine learning unit 24 can set a stop word to exclude part of the crawler data that contains the stop word from the similar comparison. , for example, "fund" is a word commonly included in fund names. Similar comparisons with the word "fund" can easily lead to inaccurate judgments. The second machine learning unit 24 can set "fund" as a stop word. , and excludes the word "fund" in the crawler data from the similar comparison, when the second machine learning unit 24 performs a comparison of the two fund names "Allianz Income Growth Fund" and "AllianzBernstein-Emerging Markets Fund" During the similar comparison, the second machine learning unit 24 removes the strings involving the stop words in the names of the two funds in the similar comparison, and uses the names of "Allianz Income Growth" and "AllianzBernstein-Emerging Markets" The names of the two funds are compared for similarity.

The client device 30 may include a display unit 31 and an input unit 32. When the client device 30 is connected to the search page 211 of the processing unit 21 through the network, the client device 30 uses the display unit 31 to The search page 211 is presented, and the search string is input in the search page 211 through the input unit 32 .

Please refer to Figure 3. The processing unit 21 is provided with a weighting page 213. The weighting page 213 is used to update and manage the weighting data. The administrator can input a new weighting column through an input field 214 of the weighting page 213. , and sets or changes the value of a weight of the weighted column, the weighting unit updates the weighted data based on the new weighted column and the weight, and transmits the new weighted data to the storage device 10 for storage.

The plurality weighted column can include single subscription, total number of favorites added, fixed subscription, single fund size, fixed fund size, number of single subscriptions, fund code, number of fixed subscriptions, fund name, number of clicks, etc., among which, fund The weighted column of code and fund name refers to the order of fund code and fund name in each fund information.

When the processing unit 21 determines that the fund name keyword data contains plural fund names corresponding to the search string, and retrieves plural fund data corresponding to the plural fund names from the storage device 10, the processing unit 21 determines according to The weighted data sorts the plural fund data. For example, if the weighted columns corresponding to the plural fund data are single subscriptions with a weight of 20 and fixed-amount subscriptions with a weight of 19, then the processing unit 21 will use the weighted column to compare the multiple fund data. The information on the two funds is sorted by the weighted column that can be used for single subscription.

Referring further to Figures 4 to 6, the processing unit 21 can set a special weight in the weighted data through the weighting page 213. The special weight can include a plurality of tags, each tag corresponding to at least one fund product, and the manager can According to different sales strategies or product plans, a corresponding label is set in the special weight, thereby changing the product ranking of each fund product in the plurality of labels displayed on the search page 211 through the special weight, and the administrator can By increasing the weight of the special weight, the ranking of the fund products to be promoted in each search result is improved. Conversely, the manager also decreases the ranking of non-mainly promoted fund products by lowering the weight of the special weight. The order in which each search result is ranked.

For example, when the manager selects fund products suitable for retirees as the main sales force, the manager adds a label named "Haoxiang Retirement" to the special weight through the weighting page 213, and adds a label named "Haoxiang Retirement" to the special weight. A plurality of fund products are set under the tag "Hao Xiang Retreat". When the client device 30 searches for fund products, in each search result, the processing unit 21 adjusts the "Hao Xiang Retreat" according to the special weight in the weighted data. The order of fund products under the label "Quick Refund" is changed to change the order in which the fund products under the label "Enjoy Refund" are displayed on the search page 211. Similarly, in each similar search result, the first machine learning unit 22 adjusts the ranking of the fund products under the "Haoxianghui" label according to the special weight in the weighted data, so as to change the order of the fund products belonging to the "Haoxianghui" label. The fund products under the tag are displayed in the order on the search page 211.

Please refer to Figures 7 and 8. In addition to the search field 212 for the client device 30 to input the search string for fund search, the search page 211 is further provided with a plurality of keyword options 215. and a plurality of popular search keywords, the plurality of keyword options 215 correspond to the fund name keyword data, and the plurality of popular search keywords correspond to each of the keywords in the fund name keyword data whose search times are greater than a preset value, The plurality of keyword options 215 and the plurality of popular search keywords are used as keyword references when the user searches for funds, and the user is helped to understand what keywords can be used to search for funds, thereby improving the speed of the user's search for funds.

Please refer to FIG. 9 . When the processing unit 21 displays the search results or the similar search results on the search page 211 , the processing unit 21 displays various items on the search page 211 according to the search results or the similar search results. The fund product's fund code, fund name, risk rating and other information, and the client device 30 can add each fund product to a shopping cart list on the search page 211, or directly on the search page 211 Purchase various fund products in 211.

Referring to FIG. 10 , the following is an example of a process in which the first machine learning unit 22 performs similarity comparison analysis on the search string and the fund name keyword data. When the client device 30 inputs a search string for "Tong Bo Mai" through the search field 212 in the search page 211, the processing unit 21 first determines whether the fund name keyword information contains the corresponding keyword. The at least one fund name or keyword of the search string. However, the fund name keyword data does not include any keyword or fund name corresponding to "Tong Bo Mao", so the processing unit 21 converts the search word The string is transmitted to the first machine learning unit 22 for further similarity comparison analysis.

The first machine learning unit 22 identifies, based on the fund name keyword data, at least one fund name whose fund name or keyword is similar to the search string. relationship, it is determined that "Tongbomai" is similar to a keyword "Lubomax" in the keyword name of the fund name. Therefore, the first machine learning unit 22 retrieves the corresponding keyword "Lubomax" from the storage device 10 At least one fund information of at least one fund name of "Mai", and then the at least one fund information is sorted according to the weighted information to generate a similar search result, and the similar search result is transmitted to the processing unit 21, and the processing unit 21 The similar search results are presented through the search page 211.

Referring to FIG. 11 , another example is used to illustrate the process of the first machine learning unit 22 performing similar comparison analysis on the search string and the fund name keyword data. When the client device 30 inputs a search string of "ㄅㄟㄌㄞˊㄉㄜ" through the search field 212 in the search page 211, the processing unit 21 first determines whether the fund name keyword information contains Contains the at least one fund name or keyword corresponding to the search string. However, the fund name keyword data does not contain any keyword or fund name corresponding to "ㄅㄟㄌㄞˊㄉㄜ", so The processing unit 21 transmits the search string to the first machine learning unit 22 for further similarity comparison analysis.

The first machine learning unit 22 identifies, based on the fund name keyword data, at least one fund name whose fund name or keyword is similar to the search string. Based on the partial phonetic-pinyin relationship, it is determined that "ㄅㄟㄌㄞˊㄉㄜ" is similar to a keyword "BlackRock" in the keyword name of the fund name. Therefore, the first machine learning unit 22 retrieves data from the storage device 10 Obtain at least one fund information corresponding to at least one fund name of the keyword "BlackRock", then sort the at least one fund information according to the weighted information to generate a similar search result, and transmit the similar search result to the processing unit 21. The processing unit 21 presents the similar search results through the search page 211.

To sum up, in the new fund search optimization system, the processing unit 21 first compares the search string entered when the client device 30 performs a fund search with the fund name keyword data, and determines whether there is any match with the search. At least one fund product corresponding to the string, and a search result is generated based on the at least one fund product, and when the search string input by the client device 30 cannot directly correspond to the fund name key due to differences in pinyin or word selection When the fund name or keyword in the word data is used, the first machine learning unit 22 further performs a similarity comparison between the search word string and the fund name keyword data, and then based on at least one fund product that is similar to the search word string. A similar search result is generated for user reference. On the other hand, the second machine learning unit 24 can continuously compare the fund names of other fund websites with the fund name keyword data according to the crawler unit 23, and update them in real time. Keyword information of the fund name. Compared with the conventional technology, this new method not only presents accurate search results to customers, but also presents similar search results to customers when the search string cannot directly correspond to the keyword information of the fund name, which helps to improve the fund name. The benchmark degree of search and increase the chance of users searching for intended products.

10:Storage device 20:Server 21: Processing unit 22: The first machine learning unit 23:Crawler unit 24:Second Machine Learning Unit 30: Client device 31:Display unit 32:Input unit 211:Search page 212:Search field 213: Weighted page 214:Input field 215:Keyword options

Figure 1: System block diagram of this new fund search and optimization system. Figure 2: A schematic diagram of the search page showing the preset display results in this new model. Figure 3: Schematic diagram of the first page of the weighted page in this new model. Figure 4: Schematic diagram of the second page of the weighted page in this new model. Figure 5: Schematic diagram of the third page of the weighted page in this new model. Figure 6: Schematic diagram of the fourth page of the weighted page in this new model. Figure 7: A schematic diagram of the search page displaying popular search keywords in this new model. Figure 8: A schematic diagram of the search page showing popular search keywords and keyword options in this new model. Figure 9: Schematic diagram of the search page displaying the fund list in this new model. Figure 10: Schematic diagram of the first page showing similar search results on the search page in this new model. Figure 11: A schematic diagram of the second page showing similar search results on the search page in this new model.

10:Storage device

20:Server

21: Processing unit

22: The first machine learning unit

23:Crawler unit

24:Second Machine Learning Unit

30: Client device

31:Display unit

32:Input unit

Claims (8)

A fund search and optimization system receives a search string transmitted by a client device. The fund search and optimization system includes: A storage device that stores a plurality of fund information and a fund name keyword data. The fund name keyword data includes the fund name of each fund product and at least one corresponding keyword; A server electrically connected to the storage device, including: A processing unit, electrically connected to the storage device, the processing unit determines whether there is at least one fund name or at least one keyword corresponding to the search string in the fund name keyword data. When the processing unit determines that the fund name keyword data When the at least one fund name or the at least one keyword corresponding to the search string is included, a search result is generated based on the at least one fund information corresponding to the at least one fund name or the at least one keyword, and the search result is Results are presented via a search page; and A first machine learning unit is electrically connected to the storage device and the processing unit. When the processing unit determines that the fund name keyword information does not include the at least one fund name or the at least one keyword corresponding to the search string, The processing unit provides the search string to the first machine learning unit; the first machine learning unit analyzes the fund name keyword information and at least one fund name or at least one keyword, and based on the search string similar to the At least one fund name or at least one fund information corresponding to the at least one keyword generates a similar search result; the first machine learning unit transmits the similar search result to the processing unit, and the processing unit passes the similar search result through The search page is rendered. For the fund search optimization system described in request item 1, the server further includes: A crawler unit searches the fund names of various fund products on other fund websites through the Internet and generates crawler data; A second machine learning unit is electrically connected to the storage device and the crawler unit. The second machine learning unit receives the crawler data and performs a similarity comparison between the crawler data and the fund name keyword data. When judging the crawler data When any fund name in the fund name keyword data is similar to the fund name or keyword, the second machine learning unit sets a new keyword based on the fund name in the crawler data, and The fund name keyword data is updated according to the new keyword, and then the new fund name keyword data is transmitted to the storage device for storage. The fund search optimization system as described in claim 1, wherein the storage device stores weighted data; The processing unit generates the search result after sorting the at least one fund information corresponding to the at least one fund name or the at least one keyword according to the weighted data. The fund search optimization system as described in claim 3, wherein before receiving the search string, the processing unit sorts the plurality of fund data according to the weighted data to generate a default display result, and presents the default display result in this search page. The fund search optimization system as described in claim 1, wherein the storage device stores weighted data; The first machine learning unit sorts the at least one fund name or the at least one fund information corresponding to the at least one keyword that is similar to the search string according to the weighted data and generates the similar search results. The fund search optimization system as described in request item 3 or 5, wherein the weighted data includes a plurality of weighted columns and a weight of each weighted column, and the processing unit is provided with a weighted page, and the weighted page is used to set and update the plurality of weighted columns. The weighted columns and the weight of each weighted column. The fund search optimization system as described in claim 6, wherein the plural weighted column includes a special weight, the special weight includes plural tags, each tag corresponds to a plurality of fund products, and the processing unit changes the plurality of tags according to the special weight. The order of products displayed on the search page for each fund product. The fund search optimization system as described in claim 2, wherein the second machine learning unit sets a stop word to exclude words containing the stop word in the crawler data from similar comparison.

Images