Program search device

Abstract

To provide a program search device that can accurately search for programs even when user queries are ambiguous. [Solution] The search query generation unit 12 generates diverse search queries based on the user query, taking into account the user's intent. The search vector generation unit 13 generates a search vector by embedding the search query into a vector. The vector search unit 14 searches the program vector DB 22 for programs that have program vectors similar to the search vector. The relevance evaluation unit 15 evaluates the degree of relevance between the programs found by the vector search unit 14 and the user query. The program search device 1 displays the programs found by the vector search unit 14 along with the degree of relevance.

Description

[Technical Field]

[0001] The present invention relates to a program search apparatus with which a user searches for a program the user desires to view. [Background Art]

[0002] Conventionally, various program search apparatuses for a user to search for a program the user desires to view have been proposed. These program search apparatuses basically operate such that a user inputs keywords relating to a program the user desires to view into the program search apparatus. [Prior Art Document] [Patent Document]

[0003] [Patent Document 1] Japanese Unexamined Patent Application Publication No. 2012-18615 [Summary of the Invention] [Problem to be Solved by the Invention]

[0004] However, it is not easy to express a user's detailed requirements with keywords. Additionally, due to orthographic variations in keywords, a user needs to undergo trial and error during searching. Furthermore, each method is specialized for a specific type of search and is not versatile. That is, it is necessary to introduce different specialized methods for (a) searching based on a user query, (b) searching for similar programs to a specific program, (c) searching for program recommendation using a user's viewing history, and the like, resulting in complicated database management.

[0005] An object of the present invention is to provide a program search apparatus that can accurately search for a program even when a user query is ambiguous. [Means for Solving the Problem]

[0006] (1) The first program search device according to the present invention includes a program data DB that stores program data comprising a program ID, a program title, and a summary text; a program vector DB that stores program vectors, which are created by embedding text concatenating the program title and summary text into a vector, in pairs with the program ID; a search query generation unit that generates diverse search queries based on user queries, taking into account the user's intent; a search vector generation unit that generates search vectors by embedding the search queries into vectors; and a vector search unit that searches the program vector DB for programs having program vectors similar to the search vectors, and displays the program data of the programs found by the vector search unit.

[0007] (2) The first program search device according to the present invention further comprises a relevance evaluation unit that evaluates the degree of relevance between the program search results of the vector search unit and the user query, and the program data of the program search results of the vector search unit may be displayed with a degree of relevance.

[0008] (3) The first program search device according to the present invention may further include a recommendation reason generation unit that generates reasons for recommending a program from the search results of the vector search unit.

[0009] (4) The first program search device according to the present invention may further include a program selection unit that allows the user to select programs to include in a playlist from the programs of the search results of the vector search unit, which are displayed with a degree of relevance.

[0010] (5) The first program search device according to the present invention may further include a title and summary generation unit that creates a humorous title and summary that will catch the user's eye for the program selected by the program selection unit.

[0011] (6) The second program search device according to the present invention includes a program data DB that stores program data comprising a program ID, a program title, and a summary; a program vector DB that stores program vectors, which are text concatenating the program title and summary embedded in a vector, in pairs with the program ID; a viewing history DB that stores the program IDs, program titles, and summary texts of programs that the user has watched; a viewing history search unit that searches the program vector DB for programs having program vectors similar to the program the user is currently watching, and extracts only programs that the user has watched from the search results; a user query generation unit that creates a user query based on the extraction results of the viewing history search unit; a search query generation unit that generates a diverse search query that takes into account the user's intent based on the user query; a search vector generation unit that generates a search vector in which the search query is embedded in a vector; and a vector search unit that searches the program vector DB for programs having program vectors similar to the search vector, and displays the program data of the programs found in the search results of the vector search unit.

[0012] (7) The second program search device according to the present invention further comprises a relevance evaluation unit that evaluates the degree of relevance between the program in the search results of the vector search unit and the user query, and the program data of the program in the search results of the vector search unit may be displayed with a degree of relevance. [Effects of the Invention]

[0013] According to the present invention, even if the user query is ambiguous, programs can be searched with high accuracy. [Brief explanation of the drawing]

[0014] [Figure 1] This diagram shows the configuration of the program search device 1 according to Embodiment 1 of the present invention. [Figure 2] This figure shows the hardware configuration of the program search device 1 according to Embodiment 1 of the present invention. [Figure 3] This diagram shows the operation of the program vector generation unit 11 of the program search device 1 according to Embodiment 1 of the present invention. [Figure 4]FIG. 1 is a diagram illustrating an operation of a search query generating unit of a program search device 1 according to the first embodiment of the present invention. [Figure 5] FIG. 1 is a diagram illustrating an operation of a search vector generating unit of the program search device 1 according to the first embodiment of the present invention. [Figure 6] FIG. 1 is a diagram illustrating an operation of a vector search unit of the program search device 1 according to the first embodiment of the present invention. [Figure 7] FIG. 1 is a diagram illustrating an operation of a relevance evaluation unit of the program search device 1 according to the first embodiment of the present invention. [Figure 8] FIG. 1 is a diagram illustrating a comparison between Bi-Encoder and Cross-Encoder. [Figure 9] FIG. 1 is a diagram illustrating an operation of a recommendation reason generating unit of the program search device 1 according to the first embodiment of the present invention. [Figure 10] FIG. 2 is a diagram illustrating a configuration of a program search device 2 according to a second embodiment of the present invention. [Figure 11] FIG. 2 is a diagram illustrating an operation of a search query generating unit of the program search device 2 according to the second embodiment of the present invention. [Figure 12] FIG. 2 is a diagram illustrating an operation of a program selecting unit of the program search device 2 according to the second embodiment of the present invention. [Figure 13] FIG. 2 is a diagram illustrating an operation of a title / summary sentence generating unit of the program search device 2 according to the second embodiment of the present invention. [Figure 14] FIG. 3 is a diagram illustrating a configuration of a program search device 3 according to a third embodiment of the present invention. [Figure 15] FIG. 3 is a diagram illustrating an operation of a viewing history search unit of the program search device 3 according to the third embodiment of the present invention. [Figure 16] FIG. 3 is a diagram illustrating an operation of a user query generating unit of the program search device 3 according to the third embodiment of the present invention. DESCRIPTION OF EMBODIMENTS

[0015] Hereinafter, embodiments of the present invention will be described with reference to the drawings. [First Embodiment] FIG. 1 is a diagram illustrating a configuration of the program search device 1 according to the first embodiment of the present invention. The program search device 1 includes a program data DB (database) 21, a program vector DB (database) 22, a program vector generation unit 11, a search query generation unit 12, a search vector generation unit 13, a vector search unit 14, a relevance evaluation unit 15, a recommendation reason generation unit 16, and an LLM (Large Language Models) 17.

[0016] The program data DB21 stores the program ID (Identification), program title, and program summary for each program. The program ID, title, and summary together are called "program data." The program vector DB22 stores a program vector for each program, which is a vector containing the program title and summary.

[0017] The program vector generation unit 11 generates a program vector in which the program title and summary are embedded in the vector. The search query generation unit 12 generates a specific search query for searching for the program the user is looking for, based on the user query issued by the user. In this case, multiple search queries may be generated for a single user query.

[0018] The search vector generation unit 13 generates a search vector by embedding the search query generated by the search query generation unit 12 into the vector. The vector search unit 14 searches the program vector DB 22 for programs that have a program vector similar to the search vector generated by the search vector generation unit 13. In general, there may be multiple programs in this search result for a single search vector.

[0019] The relevance evaluation unit 15 calculates the degree of relevance between the program in the search results of the vector search unit 14 and the user query. The evaluation result of the relevance evaluation unit 15 is called the relevance score. The relevance score can be a numerical value between 0 and 1.0. The recommendation reason generation unit 16 generates reasons why a program with a high relevance score is considered appropriate for the user query.

[0020] LLM17 is a neural network trained using a large amount of text data, capable of understanding context and handling various language-related tasks such as translation, question answering, and text creation. Note that LLM17 does not need to be installed inside the program search device 1; it may be installed outside the program search device 1 and accessed via the communication unit provided by the program search device 1 (described later).

[0021] Figure 2 shows the hardware configuration 100 of the program search device 1 according to Embodiment 1 of the present invention. The hardware configuration 100 of the program search device 1 includes a CPU (Central Processing Unit) 101, a ROM (Read Only Memory) 102, a RAM (Random access memory) 103, a storage unit 104, an input unit 105, an output unit 106, a communication unit 107, and a bus 108. The CPU 101, ROM 102, RAM 103, storage unit 104, input unit 105, output unit 106, and communication unit 107 are interconnected via the bus 108.

[0022] The CPU 101 operates according to the control program stored in the ROM 102. The ROM 102 is composed of a non-volatile semiconductor memory that is read-only and stores the control program, which is a computer program for executing processing in the program search device 1. The RAM 103 is composed of a volatile semiconductor memory that is read-and-write and provides a work area for when the CPU 101 executes a program.

[0023] The storage unit 104 is composed of, for example, a hard disk unit. The input unit 105 writes data to the storage unit 104. The output unit 106 reads data stored in the storage unit 104. The communication unit 107 exchanges data with other devices.

[0024] The CPU 101 operates using the RAM 103 as a work area, in accordance with the control program stored in the ROM 102. Functionally, the CPU 101, ROM 102, RAM 103, and storage unit 104 constitute the program data DB 21, program vector DB 22, program vector generation unit 11, search query generation unit 12, search vector generation unit 13, vector search unit 14, relevance evaluation unit 15, recommendation reason generation unit 16, and LLM 17 shown in Figure 1.

[0025] Figure 3 shows the operation of the program vector generation unit 11 of the program search device 1. The program vector generation unit 11 retrieves program data for a single program from the program data DB 21. The program data consists of a program ID, title, and summary. The program vector generation unit 11 embeds the concatenated text of the title and summary into a vector, which is a numerical representation that captures its meaning. For this vector embedding, a model trained to capture the semantic representation of text can be used, such as a Bi-Encoder type model pre-trained on a large amount of text data through comparative learning. The vector generated in this way is called a "program vector".

[0026] The program vector generation unit 11 registers the program ID and program vector as a pair in the program vector database 22. By managing the program ID and program vector as a pair, the original program data can be referenced from the program vector based on the program ID during a program search. The program vector generation unit 11 performs this series of processes in advance for each program stored in the program data database 21.

[0027] Figure 4 shows the operation of the search query generation unit 12 of the program search device 1. For example, suppose the user query is "sports other than baseball". The search query generation unit 12 creates a search query generation prompt to be input to the LLM 17. For example, the search query generation prompt may be: "Understanding the intent of the following user query, Please output a specific search query for the program the user is looking for. Please generate five search queries, each of which should be diverse and varied. #User queries "Sports programs other than baseball" The text is like this.

[0028] LLM17 can understand the context of input text and handle various language-related tasks such as translation, question answering, and text creation, for example, "Soccer broadcast" • Basketball game footage Watching a martial arts match • Tennis tournament highlights "Broadcast of golf tournaments" It outputs specific answers like the following.

[0029] Figure 5 shows the operation of the search vector generation unit 13 of the program search device 1. The search vector generation unit 13 embeds the search query into a vector using the same method as the program vector generation unit 11. The vector generated in this way is called a "search vector".

[0030] Figure 6 shows the operation of the vector search unit 14 of the program search device 1. The vector search unit 14 searches the program vector DB 22 for each search vector and extracts programs that have program vectors with a high cosine similarity to the search vector. "Cosine similarity" is the cosine of the angle between two vectors. The smaller the angle between the two vectors, the larger the cosine similarity value, which is closer to 1. Multiple similar programs are extracted for each search query, and the vector search unit 14 merges the search results by keeping only the program with the highest cosine similarity among programs with the same program ID and deleting the others. Alternatively, the cosine similarity between the search vector and the program vector calculated by the vector search unit 14 may be used instead of the relevance score output by the relevance evaluation unit 15.

[0031] Figure 7 shows the operation of the relevance evaluation unit 15 of the program search device 1. The relevance evaluation unit 15 retrieves program data from the program data DB 21 based on the program IDs from the processing results of the vector search unit 14. The relevance evaluation unit 15 creates program text by concatenating the title and summary of the program data. The relevance evaluation unit 15 uses a Rerank model to evaluate the relevance between the program text and the user query "sports programs other than baseball." The Rerank model is a ranking technology used in search results and information retrieval systems to re-evaluate the initially generated results and re-rank them to obtain more appropriate results.

[0032] To implement the Rerank model, there are three specific methods that can be considered. (1) Use Cross-Encoder. (2) Use a Cross-Encoder that has been retrained based on user evaluations. (3) Use LLM. The method described in (2) above involves showing the relevance score output by the Cross-Encoder to a human, who then evaluates it, and using the Cross-Encoder retrained based on the human evaluation results.

[0033] Figure 8 shows a comparison between bi-encoders and cross-encoders. Bi-Encoder calculates cosine similarity after creating feature vectors for two sentences, A and B. Since the cosine of the angle between two vectors takes a value between -1 and +1, the cosine similarity also takes a value between -1 and +1. Bi-Encoder can create feature vectors for each sentence even if the combination of sentences to calculate similarity for is not yet determined, so it can handle a large number of sentences by creating feature vectors in advance.

[0034] Cross-Encoder processes two sentences, A and B, using BERT, and then processes them with a binary classifier. The relevance score takes a value from 0 to +1. In Cross-Encoder, the calculation of the similarity between the two sentences is performed by the AI ​​model itself. In Cross-Encoder, if there are many sentences, the initial BERT processing cannot even begin until the combinations of which sentences to calculate the similarity between have been determined.

[0035] In Figure 7, the program with program ID "9001032" has a relevance score of "0.98". Since the user query is "sports programs other than baseball", it is reasonable that the program "9001032", which is a program about Olympic soccer, has a relevance score close to 1. In Figure 7, the program with program ID "4501024" has a relevance of "0.12". Although the program "4501024" contains the keyword "fighting," this program is ultimately a game and not a sport in which humans exercise, so it is natural that its relevance to the user query "sports programs other than baseball" is low.

[0036] Figure 9 shows the operation of the recommendation reason generation unit 16 of the program search device 1. The recommendation reason generation unit 16 sorts the programs by their relevance score. The recommendation reason generation unit 16 extracts the top k programs (where k is an arbitrary natural number) and creates a prompt for generating recommendation reasons.

[0037] The prompt for generating recommendation reasons is, for example, "Please determine if the programs listed below meet the user's requirements and output the reasons for recommending each program." #User queries "Sports programs other than baseball" #program Program ID: 9001032 Title: Paris Olympics 2024 ◇Men's Football Final: France vs. Spain, SS, Subtitles Summary: ▼Broadcast begins at 0:40 AM on NHK General TV▼The ultimate battle! World No. 2, host country France; World No. 3, Spain. Who will win the gold medal? (abbreviated) That is the case.

[0038] The recommendation reason generation unit 16 inputs the generated recommendation reason generation prompt to the LLM 17. The LLM 17 understands the context of the input text and can process various language-related tasks such as translation, question answering, and text creation. For example, "Program ID: 9001032" Relevance score: 0.98 Reason for recommendation: This program broadcast the Olympic football final, which matches the user's needs. Outputs.

[0039] The program search device 1 presents the top k programs to the user with reasons for recommendation. Note that the recommendation reason generation unit 16 is not mandatory; the program search device 1 may display program data in order of relevance score without providing recommendation reasons.

[0040] If, hypothetically, the search query generation unit 12 were removed from the program search device 1 and the user query were directly input to the search vector generation unit 13, since the user query contains the keyword "baseball," it can be expected that when the vector search unit 14 performs a search based on cosine similarity, the output of the vector search unit 14 will be programs related to baseball, such as "high school baseball at Koshien Stadium." Therefore, there is a reason to keep the search query generation unit 12.

[0041] In the program search device 1 of Embodiment 1 of the present invention, the search query generation unit 12 generates a search query that captures the user's intent based on the user query. Therefore, even if the user query is ambiguous (for example, "sports other than baseball"), it can be expanded into a specific search query suitable for the search (for example, soccer, basketball, tennis), allowing for a search. Thus, variations in keyword notation can be absorbed.

[0042] Furthermore, in the program search device 1 of Embodiment 1 of the present invention, the vector search unit 14 extracts programs that have program vectors close to the search vector, and then the relevance evaluation unit 15 evaluates the relevance score with the user query. As a result, programs that are close to the search query in terms of vectors (i.e., keywords) but actually have a low relevance to the user query are placed lower in the search results, and programs that meet the user's needs are displayed higher in the search results.

[0043] [Embodiment 2] Figure 10 shows the configuration of the program search device 2 according to Embodiment 2 of the present invention. The program search device 2 includes a program selection unit 32 and a title / summary generation unit 33, instead of the recommendation reason generation unit 16 of the program search device 1. The program selection unit 32 accepts program selections from the user while displaying program data in order of relevance score. The title / summary generation unit 33 generates humorous titles and summaries that will catch the user's eye for the programs selected by the program selection unit 32.

[0044] Figure 11 shows the operation of the search query generation unit 12 of the program search device 2. The search query generation unit 12 in Figure 11 is the same as the search query generation unit 12 in Embodiment 1, but the user query has been changed to "a program introducing local ramen from the three Tokai prefectures." Therefore, a search query such as "a special feature on ramen from the Tokai region" is generated. When searching for programs to add to an existing playlist, it is also possible to enter the title and description of the existing playlist instead of a user query. In this case, the search query generation unit 12 will generate a search query based on the title and description of the existing playlist.

[0045] Figure 12 shows the operation of the program selection unit 32 of the program search device 2. With program data displayed in order of relevance score, the device accepts user checks on the displayed rectangular blocks.

[0046] Figure 13 shows the operation of the title / summary text generation unit 33 of the program search device 2. For the program selected by the program selection unit 32, the title / summary text generation unit 33 generates a prompt to the LLM 17, for example, "Please output the titles and descriptions of the following program playlists. The titles should be humorous and eye-catching for the user." Create the file and input it into LLM17 along with the user query, program title, and summary text.

[0047] LLM17 is, for example, Title: This is the best ramen in Tokai! Summary: A compilation of TV programs showcasing delicious ramen from the three Tokai prefectures! You're sure to find a shop that catches your eye! Outputs.

[0048] In the program selection device 2 of Embodiment 2 of the present invention, the program selection unit 32 directly reflects the user's preferences, making it possible to exclude programs that do not suit the user's wishes.

[0049] [Embodiment 3] Figure 14 shows the configuration of the program selection device 3 according to Embodiment 3 of the present invention. The program selection device 3 newly includes a viewing history DB 43, a viewing history search unit 41, and a user query generation unit 42. The viewing history DB 43 stores program data of programs that the user has watched in the past. The viewing history search unit 41 extracts programs similar to the program the user is currently watching from the program vector DB 22, and then extracts only the programs that the user has already watched from those similar programs. The user query generation unit 42 generates a user query based on the program data of the programs output by the viewing history search unit 41 and the program data of the program the user is currently watching.

[0050] Figure 15 shows the operation of the viewing history search unit 41 of the program search device 3. The viewing history search unit 41 concatenates the title and summary of the program the user is currently watching (hereinafter referred to as "the program being watched") to create program text. The viewing history search unit 41 embeds the program text into a vector to create a search vector. The viewing history search unit 41 searches the program vector DB using the search vector and obtains the program IDs of programs similar to the program being watched (hereinafter referred to as "similar programs"). The viewing history search unit 41 extracts only the programs that have been stored in the viewing history DB 43 (hereinafter referred to as "similar programs that have been watched") from the similar programs.

[0051] Figure 16 shows the operation of the user query generation unit 42 of the program search device 3. The user query generation unit 42 provides, for example, a prompt for generating user queries: "The user is currently watching the following programs: Additionally, the user has previously watched the following programs: Based on this information, estimate the user's interests and generate queries to search for programs that should be presented as related programs to the program they are currently watching. This is generated and entered into LLM17 along with the title and summary of the program currently being watched, as well as the titles and summaries of similar programs that have been watched.

[0052] LLM17 infers that "the user wants information about ramen in the Tokai region," and uses this as a user query, for example: "I'd like to know about TV programs that feature delicious ramen, especially those focusing on restaurants in the Tokai region." Outputs.

[0053] In the program selection device 3 of Embodiment 3 of the present invention, a user query is created based on similar programs that have been viewed and programs currently being viewed. Therefore, even if a user query is not directly provided, a list of programs that the user may be interested in can be generated.

[0054] Although embodiments 1 to 3 of the present invention have been described above, the present invention is not limited to the embodiments described above. Furthermore, the effects described in these embodiments are merely a list of the most preferred effects arising from the present invention, and the effects of the present invention are not limited to those described in these embodiments.

[0055] In this embodiment, the configuration and operation of the program search device have been described, but the present invention is not limited thereto, and may be configured as a method or program for searching for programs, comprising each of the components.

[0056] Furthermore, this may be achieved by recording a program for realizing the functions of a program search device on a computer-readable recording medium, and then having a computer system read and execute the program recorded on this recording medium.

[0057] In this context, "computer system" includes hardware such as the operating system and peripheral devices. Furthermore, "computer-readable recording media" refers to portable media such as flexible disks, magneto-optical disks, ROMs, and CD-ROMs, as well as storage devices such as hard disks built into computer systems.

[0058] Furthermore, "computer-readable recording media" may include those that dynamically hold programs for a short period of time, such as communication lines used when transmitting programs via networks such as the Internet or communication lines such as telephone lines, as well as those that hold programs for a fixed period of time, such as volatile memory within computer systems that act as servers or clients in such cases. In addition, the above-mentioned program may be for the purpose of realizing some of the functions described above, and may also be a program that can realize the above-mentioned functions in combination with a program already recorded in the computer system. [Explanation of symbols]

[0059] 1. Program search device 2. Program search device 3. Program search device 11 Program Vector Generation Unit 12 Search Query Generation Unit 13 Search Vector Generation Unit 14. Vector Search Unit 15. Relevance Evaluation Department 16 Recommendation reason generator 17 LLM 21 Program Data Database 22 Program Vector Database 32 Program Selection Section 33 Title / Summary Generation Unit 41 Viewing History Search Section 42 User Query Generation Unit 43 Viewing History Database 100. Hardware configuration of the program search device 101 CPU 102 ROM 103 RAM 104 Storage section 105 Input section 106 Output section 107 Communications Department 108 Bus

Claims

1. A program data database containing program data including program ID, program title, and summary text, A program vector database stores program vectors, which are text vectors containing the program title and summary, paired with the program ID. A search query generation unit that understands the user's intent and generates diverse search queries based on user queries, A search vector generation unit generates a search vector in which the aforementioned search query is embedded in a vector, The system includes a vector search unit that searches the program vector DB for programs having program vectors similar to the search vector, A program search device that displays program data for programs found in the search results of the vector search unit.

2. The system further includes a relevance evaluation unit that evaluates the degree of relevance between the program results of the vector search unit and the user query. The program search device according to claim 1, wherein the program data of the search results of the vector search unit is displayed with a degree of relevance.

3. The program search device according to claim 1, further comprising a recommendation reason generation unit that generates reasons for recommending programs from the search results of the vector search unit.

4. The program search device according to claim 1, further comprising a program selection unit that allows the user to select programs to include in a playlist from the search results of the vector search unit, which are displayed with a degree of relevance.

5. The program search device according to claim 4, further comprising a title and summary generation unit that creates a humorous title and summary that will catch the user's eye for the program selected by the program selection unit.

6. A program data database containing program data including program ID, program title, and summary text, A program vector database stores program vectors, which are text vectors containing the program title and summary, paired with the program ID. A viewing history database that stores the program ID, program title, and summary text of programs that the user has watched, A viewing history search unit searches the program vector database for programs with program vectors similar to the program the user is currently watching, and extracts only the programs the user has already watched from the search results. A user query generation unit creates a user query based on the extraction results of the aforementioned viewing history search unit, A search query generation unit that, based on the aforementioned user query, understands the user's intent and generates diverse search queries, A search vector generation unit generates a search vector in which the aforementioned search query is embedded in a vector, The system includes a vector search unit that searches the program vector DB for programs having program vectors similar to the search vector, A program search device that displays program data for programs found in the search results of the vector search unit.

7. The system further includes a relevance evaluation unit that evaluates the degree of relevance between the program results of the vector search unit and the user query. The program search device according to claim 6, wherein the program data of the search results of the vector search unit is displayed with a degree of relevance.

8. A program data database containing program data including program ID, program title, and summary text, A program vector database stores program vectors, which are text vectors containing the program title and summary, paired with the program ID. In a device equipped with, A search query generation step that generates diverse search queries by understanding the user's intent based on the user query, A search vector generation step that generates a search vector in which the aforementioned search query is embedded in a vector, A vector search step of searching the program vector DB for programs having program vectors similar to the search vector, A program search method performed by a computer, comprising: a display step of displaying program data of the programs found in the vector search step; and a display step of displaying program data of the programs found in the vector search step.

9. The computer further performs a relevance evaluation step, which evaluates the degree of relevance between the program in the search results of the vector search step and the user query. The program search method according to claim 8, wherein in the display step, the program data of the search results of the vector search step is displayed with a degree of relevance.

10. The program search method according to claim 8, wherein the computer further performs a recommendation reason generation step in which it creates reasons for recommending the programs in the search results of the vector search step.

11. The program search method according to claim 8, wherein the computer further performs a program selection step in which the user selects programs to include in a playlist from the programs of the search results of the vector search step, which are displayed with their relevance.

12. The program search method according to claim 11, wherein the computer further performs a title and summary generation step in which it creates a humorous title and summary that will catch the user's eye for the program selected in the program selection step.

13. A program data database containing program data including program ID, program title, and summary text, A program vector database stores program vectors, which are text vectors containing the program title and summary, paired with the program ID. A viewing history database that stores the program ID, program title, and summary text of programs that the user has watched, In a device equipped with, A viewing history search step involves searching the program vector database for programs that have a program vector similar to the program the user is currently watching, and extracting only the programs the user has already watched from the search results. A user query generation step that creates a user query based on the extraction results of the aforementioned viewing history search step, A search query generation step that generates diverse search queries by understanding the user's intent based on the aforementioned user query, A search vector generation step that generates a search vector in which the aforementioned search query is embedded in a vector, A vector search step of searching the program vector DB for programs having program vectors similar to the search vector, A program search method performed by a computer, comprising: a display step of displaying program data of the programs found in the vector search step; and a display step of displaying program data of the programs found in the vector search step.

14. The computer further performs a relevance evaluation step, which evaluates the degree of relevance between the program in the search results of the vector search step and the user query. The program search method according to claim 13, wherein in the display step, the program data of the search results of the vector search step is displayed with a degree of relevance.

15. A program search program for causing a computer to function as a program search device according to any one of claims 1 to 7.

Images