Information processing system, information processing method, and program

Abstract

By accurately grasping customer needs in real time and providing appropriate property information instantly, the system improves the work efficiency of sales staff, increases the accuracy of proposals, and increases customer satisfaction. [Solution] In an information processing system that matches real estate properties, the recommendation device comprises a memory unit that stores customer property needs information and property information, a transcription result acquisition unit that acquires real-time transcription results of conversational audio with the customer, a property needs information acquisition unit that acquires customer property needs information automatically extracted by AI analysis based on the transcribed conversation content, a property extraction unit that extracts properties that match the customer property needs information from the property information stored in the memory unit based on the customer's property needs, a reason acquisition unit that acquires recommendation reasons for the extracted properties generated by AI analysis based on the characteristics of the property and the customer's property needs information, and an output unit that outputs recommendation results including recommendation reasons to a user terminal.

Description

[Technical Field]

[0001] The present invention relates to an information processing system, an information processing method, and a program. [Background technology]

[0002] Conventionally, real estate transaction systems have been proposed that allow tenants and landlords to access residential rental properties. For example, Patent Document 1 discloses a residential rental system in which tenants register their desired property conditions (location, layout, rent, etc.) in a database, and a search server searches a property information database for rental properties that meet the desired conditions and displays a list of the properties to the tenant. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Application Laid-Open No. 2002-32458 Summary of the Invention [Problem to be solved by the invention]

[0004] However, in the above-mentioned systems, the tenant explicitly inputs their desired conditions and the search is based on these, so if the tenant's desired conditions are unclear or if the tenant's latent needs that they cannot verbalize are not easily reflected in the search results.In addition, in order for landlords and real estate agents to understand the tenant's essential values ​​and lifestyle and propose properties that match them, they have to rely on individual responses and face-to-face interviews, which places a heavy burden on sales representatives and tends to make the quality of proposals dependent on the representative's experience and intuition.

[0005] The main purpose of the present invention is to accurately grasp customer needs in real time and provide appropriate property information immediately, thereby improving the work efficiency of sales representatives and increasing the accuracy of proposals and customer satisfaction. [Means for solving the problem]

[0006] The information processing system of the present invention comprises: An information processing system for matching real estate properties, a storage unit for storing customer property needs information and property information; a transcription result acquisition unit that acquires real-time transcription results of conversational voices with customers; a property needs information acquisition unit that acquires property needs information of the customer automatically extracted by AI analysis based on the transcribed conversation content; a property extraction unit that extracts properties that match the property needs information of the customer from the property information stored in the storage unit based on the property needs of the customer; a reason acquisition unit that acquires, for the extracted property, a recommendation reason generated by the AI ​​analysis based on the property's characteristics and the customer's property needs information; an output unit that outputs the recommendation result including the recommendation reason to a user terminal; The gist of the project is to provide the following:

[0007] According to this invention, AI analyzes and extracts customer conversation content in real time and automatically recommends the most suitable property information, eliminating the need for sales representatives to manually search for property information. This reduces human error in sales activities, significantly improves the speed and accuracy of responding to customer needs, and ultimately increases the success rate of transactions.

[0008] The information processing method of the present invention comprises: An information processing method for matching real estate properties, comprising: a storage step of storing customer property needs information and property information; a transcription result acquisition step of acquiring information transcribed in real time from a conversation voice with a customer; a property needs information acquisition step of automatically extracting property needs information of the customer from the transcribed conversation content by AI analysis and acquiring the property needs information of the customer; a property extraction step of extracting properties that meet the property needs of the customer from the property information stored in the storage step based on the property needs of the customer; a reason acquisition step of acquiring a recommendation reason automatically generated by the AI ​​analysis based on the characteristics of the extracted property and the customer's property needs; an output step of outputting the recommendation result including the recommendation reason to a user terminal; The gist of the project is to provide the following:

[0009] The information processing method of the present invention has the same effects as the information processing system of the present invention. [Brief explanation of the drawings]

[0010] [Figure 1] 1 is a schematic configuration diagram of an information processing system according to an embodiment. [Figure 2] FIG. 2 is a functional block diagram of the negotiation device according to the embodiment. [Figure 3] FIG. 2 is a functional block diagram of a recommendation device according to the embodiment. [Figure 4] FIG. 2 is an explanatory diagram illustrating an example of property information according to the embodiment. [Figure 5] 1 is a functional block diagram of a voice recognition device according to an embodiment; [Figure 6] 1 is a functional block diagram of a language processing apparatus according to an embodiment; [Figure 7] FIG. 10 is a diagram showing the first half of the flow of recommendation processing according to the embodiment. [Figure 8] FIG. 10 is a diagram illustrating the second half of the flow of the recommendation process according to the embodiment. [Figure 9] FIG. 10 is an explanatory diagram showing how property needs information is created according to the embodiment. [Figure 10] FIG. 2 is an explanatory diagram showing how a speech recognition device according to an embodiment transcribes speech. [Figure 11] FIG. 10 is an explanatory diagram showing a state in which a transcript of a conversation between a sales representative and a customer is displayed on a display device according to an embodiment. [Figure 12]FIG. 10 is an explanatory diagram showing a state in which a pop-up is displayed on the display device according to the embodiment. [Figure 13] FIG. 10 is an explanatory diagram showing a state in which a recommendation result is displayed on the display device according to the embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0011] Next, an embodiment of the present invention will be described with reference to the drawings. Fig. 1 is a schematic configuration diagram of an information processing system 1 used for matching real estate properties. The information processing system 1 includes a negotiation device 10, a recommendation device 30 connected to the negotiation device 10 via a network N, a speech recognition device 50 connected to the recommendation device 30 via the network N, and a language processing device 60 connected to the recommendation device 30 via the network N.

[0012] [Business negotiation device 10] The negotiation device 10 is a device used for negotiations between real estate salespeople and customers. The negotiation device 10 is configured as a general-purpose computer having a CPU, ROM, RAM, storage (e.g., SSD or HDD), etc. As shown in FIG. 2, the negotiation device 10 has a communication unit 11, a display device 12, a voice input device 13, an operation device 14, a memory unit 15, and a control unit 16.

[0013] The communication unit 11 is an interface for transmitting and receiving information between the negotiation device 10 and the recommendation device 30. The communication unit 11 includes, for example, a LAN interface and a wireless communication module (for example, a Wi-Fi communication module or an LTE communication module), and is capable of data communication via the network N. The communication unit 11 plays a role in transmitting and receiving various data to and from the recommendation device 30, and supports real-time processing in the negotiation device 10.

[0014] The display device 12 displays images and videos output from the control unit 16. The display device 12 is configured as, for example, a liquid crystal display or an organic EL display.

[0015] The voice input device 13 inputs voice data such as the operator's voice and environmental sounds to the control unit 16. The voice input device 13 is configured as, for example, a general-purpose microphone.

[0016] The operation device 14 is used to input operations by an operator to the control unit 16. The operation device 14 is, for example, a general-purpose device such as a mouse or a keyboard.

[0017] The storage unit 15 stores programs for executing the recommended processing described below and the various functions in the control unit 16, input data, etc., and is composed of RAM (Random Access Memory), ROM (Read Only Memory), etc.

[0018] The control unit 16 is an arithmetic processing device that is responsible for overall control of the negotiation device 10, and includes a reception unit 17, a voice storage processing unit 18, a transmission unit 19, a receiving unit 20, and a display processing unit 21. The control unit 16 comprehensively controls each functional unit by a control program executed on, for example, a CPU, and sequentially and in real time manages a series of processes from acquiring and recording the conversation voice with the customer, to sending and receiving information through communication with the recommendation device 30, and even displaying the recommendation results. The control unit 16 is also responsible for optimizing the processing timing of each unit, controlling linkage with the user interface, and flow control according to the progress of the dialogue.

[0019] The reception unit 17 receives input of information such as the sales representative's name, identifier (e.g., ID, email address, etc.), contact information, and the customer's name and contact information (e.g., telephone number, email address). The reception unit 17 acquires this information via the operation device 14 and formats it into a format that can be transmitted to the recommendation device 30. The reception unit 17 also inputs the dialogue voice during negotiations with the customer. The reception unit 17 acquires the voice input from the voice input device 13 in real time. The reception unit 17 starts acquiring the voice in response to an operation by an operator (e.g., a sales representative).

[0020] The voice storage processing unit 18 temporarily stores the input voice in the storage unit 15 .

[0021] The transmission unit 19 transmits sales representative information, which is information about sales representatives, and customer information to the recommendation device 30. The transmission unit 19 also transmits to the recommendation device 30 segmentation conditions, which are conditions used by the speech recognition device 50 when segmenting conversational voices, and importance weights used when extracting properties to be recommended to customers. The transmission unit 19 also transmits various information to the recommendation device 30, such as conversational voices stored in the storage unit 15 that are equal to or longer than a predetermined length, and conversational voices between silent periods. In this way, the transmission unit 19 adjusts the transmission timing or adjusts the transmission timing based on the length of the conversational voices.

[0022] The receiving unit 20 receives the transcription result, the recommendation result, the reason for recommendation, etc. from the recommendation device 30.

[0023] The display processing unit 21 displays the transcription results received from the recommendation device 30 on the display device 12. The transcription result display unit 22 distinguishes between utterances by speaker and displays them with good visibility. The display processing unit 21 also displays the recommendation results received from the recommendation device 30 and the reasons for them on the display device 12. The display processing unit 21 displays property information (e.g., property name, floor plan, structure, location, expected yield, etc.) 32b sorted in order of score in card format, and also lists the reasons why each property was recommended (e.g., "quiet living environment," "reinforced concrete structure with good soundproofing," etc.).

[0024] As described above, the control unit 16 receives, stores, and transmits information about sales representatives and customers, collects voice data during negotiations, and cooperates with the recommendation device 30 to present the results of recommended processing in real time, thereby improving the efficiency of sales activities and customer satisfaction. The control unit 16 also cooperatively controls each component from the reception unit 17 to the display processing unit 21, and plays a role in overseeing a seamless processing flow from the start of voice collection to display output. Furthermore, the control unit 16 dynamically processes execution decisions to improve the accuracy and immediacy of customer responses, such as adjusting the voice transmission unit and optimizing the display format.

[0025] [Recommended Device 30] The recommendation device 30 is a device that recommends real estate to the negotiation device 10. The recommendation device 30 is a general-purpose server having a CPU, ROM, RAM, storage, etc. As shown in FIG. 3 , the recommendation device 30 has a communication unit 31, a memory unit 32, a generation unit 33, a reception unit 34, a transmission unit 35, an instruction unit 36, an update unit 37, a storage processing unit 38, a setting unit 39, a transcription result acquisition unit 40, a property needs information acquisition unit 41, a property extraction unit 42, a reason acquisition unit 43, an output unit 44, an email text acquisition unit 45, an information acceptance unit 46, an energy acquisition unit 47, and a determination unit 48.

[0026] The communication unit 31 is a communication interface for transmitting and receiving information between the recommendation device 30 and the negotiation device 10, the voice recognition device 50, and the language processing device 60. The communication unit 31 transmits and receives various data bidirectionally through a wired or wireless network connection. The communication unit 31 supports, for example, multiple network ports and secure communication protocols, enabling real-time and highly reliable data linkage.

[0027] The storage unit 32 stores various data related to the recommended processing described below, such as salesperson information 32a, property information 32b, and property needs information 32c. The storage unit 32 is configured as a non-volatile storage such as an HDD or SSD. The salesperson information 32a is information including, for example, the identification information and contact information (email address and telephone number) of the salesperson.

[0028] As shown in FIG. 4, the property information 32b includes items such as area, price range, area, category, floor area ratio, structure, and expected yield for each real estate property.

[0029] The property needs information 32c is structured data including, for example, the following items: The customer's desired area (e.g., region, prefecture, city, town, etc.) Environmental risks (e.g., flood inundation forecast (area of ​​○m), landslide warning areas, noise levels, etc.) School districts and administrative services (e.g., distance to designated elementary and junior high schools, childcare facilities, etc.) Category (e.g. apartment, detached house, land, hotel, shop, etc.), Intended use (e.g., personal residence, investment management, private lodging, company housing, shared office, etc.) Year of construction (e.g., new construction, age (within x years), etc.) Earthquake resistance performance (e.g., compliance with new earthquake resistance standards, existing non-compliant, seismic control structure, seismic isolation structure, etc.) Renovation history (e.g., full renovation (year), partial renovation (year)) Facilities (e.g., elevator availability, EV charging facilities, solar power generation, smart home wiring) ·Building coverage ratio・Floor area ratio (e.g. specified upper limit (%)) Area (e.g., lower and upper bounds) · Zoning (e.g., commercial, semi-industrial, first-class residential, etc.) -Structure type (e.g., wooden, steel, reinforced concrete, steel reinforced concrete, etc.), Price range (e.g., lower limit, upper limit (10,000 yen)), - Expected yield (e.g. nominal, real (%)), Rental terms (e.g. expected rent, security deposit, key money, contract term, renewal fee) Transaction schemes (e.g., regular sales, leasebacks, condominium purchases, joint investments) CAPEX / OPEX (e.g., annual repair costs (million yen), fixed asset tax (million yen)) Future value forecast (price increase scenario (% / year)) Priority: Each item is weighted in 5 levels (Required, High, Medium, Low, Unnecessary) Comments (free text)

[0030] The generation unit 33 generates initial property needs information 32c based on the salesperson information 32a and the customer information. The generation unit 33 creates the property needs information 32c by using general conditions according to the customer's attributes, such as the customer's desired area, price range, real estate category, etc., as structured data.

[0031] The receiving unit 34 receives salesperson information, customer information, and the like from the negotiation device 10. The receiving unit 34 also receives segmentation conditions, importance weights, and the like from the negotiation device 10. The receiving unit 34 receives voice data transmitted from the negotiation device 10. The receiving unit 34 sequentially receives voice fragments segmented into predetermined units and uses them as triggers for the transcription process described below. The receiving unit 34 also receives, from the voice recognition device 50, transcription results obtained by transcribing the voice of a conversation between a salesperson and a customer, extracted information on the customer's emotions, and extracted information on the customer's enthusiasm. The receiving unit 34 also receives, from the language processing device 60, recommendation results (a group of properties to be recommended to the customer), recommendation reasons for recommending properties to the customer, updated property needs information 32c, and the text of an email to be sent to the salesperson.

[0032] The transmission unit 35 transmits the stored transcription results to the negotiation device 10 and the language processing device 60. The transcription results are used by the language processing device 60 to analyze meaning and extract needs. The transmission unit 35 also transmits recommendation results (a group of properties to be recommended to the customer) extracted by a property extraction unit 42 (described later) to the negotiation device 10 and the language processing device 60. The transmission unit 35 transmits the text of the proposal email received from the language processing device 60 to the sales representative. This allows the sales representative to quickly send a proposal email with the reason for recommendation to the customer.

[0033] Hereinafter, in this specification, instructions to update the display of recommendation results or recommendation reasons are collectively referred to as "update instructions." The instruction unit 36 ​​instructs the speech recognition device 50 to perform speech transcription processing. At this time, the instruction unit 36 ​​instructs the speech recognition device 50 to perform transcription processing, taking into account speaker separation and semantic segmentation for each utterance unit, instructs the speech recognition device 50 to perform processing to extract emotional changes based on the customer's speech tendencies, and instructs the device 60 to extract customer enthusiasm. Emotional changes include, for example, increased interest and detection of positive comments. Meanwhile, enthusiasm is an index that quantitatively evaluates the strength of interest and enthusiasm expressed in the customer's speech by comprehensively evaluating non-verbal characteristics such as the degree of emphasis, repetition, intonation, voice volume, and prolonged endings of the speech. Furthermore, the instruction unit 36 ​​instructs the language processing device 60 to perform analysis to extract property needs information 32c based on the transcription results acquired by the transcription result acquisition unit 40. The instruction unit 36 ​​also instructs the language processing device 60 to create the text of a proposal email for the customer. The instruction unit 36 ​​also instructs the negotiation device 10 to display the transcription results, to display a pop-up, or to display the recommendation results.

[0034] The update unit 37 updates the property needs information 32c stored in the storage unit 32 based on the latent needs extraction result received from the language processing device 60. The property needs information 32c is updated to reflect the customer's desired conditions, lifestyle, speech context, etc., thereby improving the accuracy of the information.

[0035] The storage processor 38 stores the transcription results sent from the voice recognition device 50. The stored character string data is used for subsequent extraction of latent needs, generation of recommendation reasons, and the like.

[0036] The setting unit 39, in response to operation of the operation device 14 of the negotiation device 10, sets a predetermined importance weight for each item included in the property needs information 32c. The importance weight may be automatically assigned using a preset default value. For example, if a customer places the highest priority on "proximity to a station," setting a high weight for the item increases the priority during matching. This weight is reflected in the similarity calculation by the property extraction unit 42. The setting unit 39 also sets segmentation conditions, which are conditions under which the speech recognition device 50 segments conversational voices, in response to operation of the operation device 14 of the negotiation device 10 connected to the recommendation device 30 via the network N. The segmentation conditions include a silent section of a predetermined duration or longer, a speech equivalent to a period at the end of a spoken sentence, or the appearance of vocabulary indicating semantic completion of the speech. The setting unit 39 can flexibly change the conditions by accepting settings from the operator and can also automatically switch setting values ​​based on profiles registered in the storage unit 32. When the setting unit 39 receives the setting of the delimiting condition, it transmits the delimiting condition to the voice recognition device 50 via the transmitting unit 35.

[0037] The transcription result acquisition unit 40 receives real-time speech recognition results transmitted from the speech recognition device 50 and acquires the transcription results. When the transcription result acquisition unit 40 detects a predetermined division condition, such as the end of an utterance or the completion of a sentence, it can divide and record the transcription results on an utterance-by-utterance basis, and the division processing conditions can be changed by the condition setting unit. This makes it possible to organize data in units suitable for needs extraction and sentiment analysis using subsequent AI analysis.

[0038] The property needs information acquisition unit 41 acquires property needs information 32c generated as an analysis result of intent estimation and keyword extraction performed by natural language processing by the language processing device 60. The property needs information acquisition unit 41 acquires the property needs information 32c as structured property needs information and stores it in the storage unit 32.

[0039] The property extraction unit 42 executes a process of selecting candidate properties using the updated property needs information 32c. The property extraction unit 42 combines two or more of the following methods to extract a group of properties with high suitability: collaborative filtering, rule-based matching, and content-based filtering.

[0040] The reason acquisition unit 43 acquires recommendation reasons based on the relevance of each piece of property information 32b extracted by the property extraction unit 42 with the customer's property needs information 32c. The recommendation reasons are generated by natural language generation processing using a large-scale language model (LLM) in the language processing device 60 (described later), and are text data that explains how the property matches the customer's desired conditions or speaking tendencies. The reason acquisition unit 43 receives and acquires natural expressions in the form of, for example, "Since the customer wanted a reinforced concrete structure, we recommend this property with excellent soundproofing," and stores them in the storage unit 32 while linking them to the property in question.

[0041] When the output unit 44 receives the property information 32b and the corresponding reason for recommendation generated by the language processing device 60, it outputs an instruction to display these at an appropriate time and in an appropriate format on the display device 12 of the negotiation device 10, which is the operating terminal of the sales representative.

[0042] Hereinafter, an email containing the proposal content sent to a customer will be referred to as a "proposal email." The email text acquisition unit 45 acquires the text of the proposal email generated based on the recommendation reasons and property needs information 32c via the receiving unit 34. The text of the proposal email is created by the AI ​​analysis unit of the language processing device 60, which will be described later. The text of the proposal email includes a natural proposal sentence, such as, "For customers who wish to live in a quiet environment, we recommend a reinforced concrete structure property in the XX area." The email text acquisition unit 45 stores the text in the storage unit 32 for confirmation and transmission by the sales representative.

[0043] When the property needs information 32c is updated, the information receiving unit 46 receives, in addition to the transcription result, at least one additional piece of information from keywords extracted from the conversational voice, emotional information about the speaker, or topic information about the conversation via the receiving unit 34. This information is provided from the speech recognition device 50 or the language processing device 60, and is intended to complementarily utilize non-structural elements contained in the customer's speech.

[0044] The energy acquisition unit 47 acquires the energy of the conversation, which is an estimate of the degree of interest (energy) when the customer makes a comment, based on the emotion information and topic information.

[0045] The determination unit 48 determines whether the heat calculated by the heat acquisition unit 47 exceeds a preset threshold. If a positive determination is made, i.e., if it is estimated that the customer's interest is sufficiently high, the output unit 44 outputs an instruction to update the display of the property or the recommended results. In this way, the system realizes dynamic display control according to the customer's reaction, and makes it possible to make timely suggestions according to the flow of the dialogue.

[0046] [Speech Recognition Device 50] The speech recognition device 50 processes the speech received from the recommendation device 30. The speech recognition device 50 is configured as, for example, a general-purpose server equipped with a speech recognition AI. As shown in FIG. 5 , the speech recognition device 50 includes a receiving unit 51, a reflecting unit 52, a transcription processing unit 53, an emotion change extracting unit 54, an energy extracting unit 55, and a transmitting unit 56.

[0047] The receiving unit 51 receives the voice data transmitted from the recommendation device 30. For example, the receiving unit 51 buffers the packetized voice stream and formats it into a format that can be transferred to the transcription processing unit 53 in fixed units. Here, the fixed unit is set by the setting unit 39 of the recommendation device 30, and is, for example, "voice per utterance," "a 5-second audio fragment," or "a segment separated by a silent interval." This allows the transcription processing unit 53, which will be described later, to perform text conversion processing while maintaining the natural unity of each utterance. The receiving unit 51 receives an instruction to extract customer emotions from the recommendation device 30.

[0048] The reflecting unit 52 reflects the delimiting conditions set by the setting unit 39 of the recommended device 30 in the processing of the transcription processing unit 53, which will be described later.

[0049] The transcription processing unit 53 analyzes the voice data input via the receiving unit 51 using deep learning-based speech recognition AI and converts the speech content contained in the voice into text data with a natural sentence structure. The transcription processing unit 53 generates highly accurate character string data that preserves the meaning of the sentence by automatically performing speaker separation, silent section detection, punctuation completion, and semantic paragraph extraction while taking into account time-series changes in acoustic features, speech intonation, and linguistic context. In addition, the speech recognition engine demonstrates robust recognition performance even in noisy environments and for a variety of speakers by integrating acoustic and language models using a neural network.

[0050] The emotion change extraction unit 54 analyzes the emotional changes of the customer contained in the conversation voice, and extracts emotion indicators such as increased interest, caution, and hesitation, for example.

[0051] The enthusiasm extraction unit 55 numerically evaluates the customer's level of interest and emotional enthusiasm based on non-verbal features such as tone of speech, emphasis, speech rate, intonation, contextual repetition, frequency of use of affirmative / negative words, laughter, sighs, etc. contained in the conversational voice. For example, the enthusiasm extraction unit 55 assigns a high enthusiasm level when emphasis expressions such as "I'm very interested" or "That's really good" are repeated. The enthusiasm extraction unit 55 also determines that the enthusiasm level is increasing when the same phrase is used repeatedly, the volume is increasing, or the ending of words is prolonged.

[0052] The transmission unit 56 transmits the character string data generated by the transcription processing unit 53 to the recommendation device 30. The transmission unit 56 sequentially transmits the data each time the transcription is completed, thereby supporting real-time processing in the recommendation device 30. The transmission unit 56 also transmits the extracted emotion change information to the recommendation device 30.

[0053] [Language Processing Unit 60] The language processing device 60 analyzes the transcription results, extracts the customer's latent needs and values, and generates the reason for recommendation and the text of the proposal email. The language processing device 60 is configured as, for example, a server that implements a large-scale language model (LLM).

[0054] As shown in FIG. 6, the language processing device 60 includes a receiving unit 61, an AI analyzing unit 62, an updating unit 63, and a transmitting unit 64.

[0055] The receiving unit 61 receives the transcription result from the recommendation device 30. The receiving unit 61 also receives the recommendation result from the recommendation device 30 and receives an instruction to generate a recommendation reason.

[0056] The AI ​​analysis unit 62 performs natural language processing using a large-scale language model on the transcription results received from the recommendation device 30, and extracts keywords, language patterns, and contextual features contained in the customer's utterance. Furthermore, based on these extraction results, the AI ​​analysis unit 62 multifacetedly estimates the customer's desired conditions, potential needs, lifestyle, values, etc. The estimation results are reflected in the generation of property needs information 32c, which is constructed as needs information optimized for each customer. The AI ​​analysis unit 62 also performs natural language generation processing using a large-scale language model on the property information 32b extracted by the recommendation processing to generate recommendation reasons that express why the property meets the customer's needs. The recommendation reasons are dynamically generated based on keywords, preference trends, emotional tone, and other factors contained in the customer's speech, and are output as natural sentences that are easy for the customer to understand and convincing. The AI ​​analysis unit 62 also performs natural language generation processing using a large-scale language model based on the property information 32b, recommendation reasons, and property needs information 32c to automatically generate a proposal email for the customer. The generated text is contextually optimized according to the customer's interests and expression style, and the recommendation content is written in a natural and persuasive manner. The text is also adjusted to emphasize the property's appeal and recommendation reasons in light of the customer's needs.

[0057] The update unit 63 updates the property needs information 32 c received from the recommendation device 30 .

[0058] The transmission unit 64 transmits the property needs information 32c generated by the AI ​​analysis unit 62 to the recommendation device 30. The transmission unit 64 transmits the generated reason for recommendation to the recommendation device 30. The transmission unit 64 transmits the text of the created proposal email to the recommendation device 30. The proposal email is ultimately provided to a sales representative. The sales representative can quickly carry out proposal activities by sending the text to the customer.

[0059] [Operation of Information Processing System 1] Next, the operation of the information processing system 1 will be described. FIG. 7 is a diagram showing the first half of the recommendation processing flow. FIG. 8 is a diagram showing the second half of the recommendation processing flow. When a sales representative starts negotiations using the negotiation device 10, a series of processes are executed sequentially, such as information input, voice acquisition, transmission, transcription and text analysis, property extraction, recommendation generation, sentiment analysis, and proposal email creation. Each step will be described in detail below.

[0060] In step S101, the reception unit 17 of the negotiation device 10 receives input of sales representative information (e.g., ID, telephone number, email address, etc.) 32a and customer information (name, telephone number, etc.) via the operation device 14. Then, the transmission unit 19 transmits the sales representative information 32a and the customer information to the recommendation device 30.

[0061] In step S102, the storage processing unit 38 of the recommendation device 30 stores the sales representative information 32a in the storage unit 32.

[0062] In step S103, the generating unit 33 of the recommending device 30 generates property needs information 32c as shown in FIG. 9(a).

[0063] In step S104, the reception unit 17 of the negotiation device 10 receives input of segmentation conditions, which are conditions for the speech recognition device 50 to segment the conversational voice, through the salesperson's operation of the operation device 14. Then, the transmission unit 19 transmits the received segmentation conditions to the recommendation device 30.

[0064] In step S105, the setting unit 39 of the recommendation device 30 receives the segmentation conditions via the receiving unit 34 and transmits the segmentation conditions via the transmitting unit 35 to the speech recognition device.

[0065] In step S106, the receiving unit 51 of the voice recognition device 50 receives the segmentation conditions. Then, the reflecting unit 52 of the voice recognition device 50 sets the received conditions as segmentation conditions when performing voice recognition.

[0066] In step S107, the reception unit 17 of the negotiation device 10 receives the input of the importance weight of each item used by the recommendation device 30 when extracting properties, along with the salesperson's operation of the operation device 14. Then, the transmission unit 19 of the negotiation device 10 transmits the received importance weight to the recommendation device 30.

[0067] In step S108, the setting unit 39 of the recommendation device 30 sets the received weight as the importance weight when extracting properties.

[0068] In step S109, the voice storage processor 18 of the negotiation device 10 starts inputting conversational voice in response to the operation of the operation device 14 by the salesperson.

[0069] In step S110, the voice storage processing unit 18 of the negotiation device 10 stores in the storage unit 15 the conversational voices between the salesperson and the customer input via the voice input device 13.

[0070] In step S111, the transmission unit 19 of the negotiation device 10 transmits a predetermined length (for example, 5 seconds) of speech to the recommendation device 30. In this way, the transmission unit 19 suppresses fragmented transmission of speech, and dynamically adjusts the timing of transmitting speech to the speech recognition device 50, targeting only meaningful units of speech data that have reached the predetermined length.

[0071] In step S112, the receiving unit 34 of the recommendation device 30 receives the conversational voice transmitted from the negotiation device 10. The transmitting unit 35 of the recommendation device 30 transmits the received conversational voice to the voice recognition device 50. Then, the instructing unit 36 ​​of the recommendation device 30 instructs the voice recognition device 50 to transcribe the conversational voice.

[0072] In step S113, the receiving unit 51 of the speech recognition device 50 receives the conversational speech from the recommendation device 30. Then, the transcription processing unit 53 of the speech recognition device 50 transcribes the conversational speech in accordance with the segmentation conditions reflected by the reflecting unit 52. Specifically, as shown in FIG. 9, the transcription processing unit 53 of the speech recognition device 50 analyzes the received conversational speech and generates text data (transcription results) while performing speaker separation and semantic segmentation. In accordance with the reflected segmentation conditions, the transcription processing unit 53 detects silence for a certain period of time (e.g., 5 seconds) or more and determines the end of a sentence, such as punctuation, to process the speech by segmenting it into meaningful speech units and organize the speech content by speaker.

[0073] In step S114, the transmission unit 56 of the voice recognition device 50 transmits the stored transcription result to the recommendation device 30.

[0074] In step S115, the transcription result acquisition unit 40 of the recommendation device 30 receives the transcription result from the speech recognition device 50 via the receiving unit 34. Then, the transcription result acquisition unit 40 of the recommendation device 30 stores the transcription result in the memory unit 32 via the storage processing unit 38.

[0075] In step S116, the transmission unit 35 of the recommendation device 30 transmits the transcription result to the negotiation device 10.

[0076] In step S117, the display processing unit 21 of the negotiation device 10 displays the transcription results on the display device 12, distinguishing between the speech of the sales representative and the speech of the customer, as shown in Fig. 10. That is, the display processing unit 21 displays the speech of the sales representative and the speech of the customer in speech bubbles on the left and right, and provides text information in a dialogue format that is updated in real time.

[0077] In step S118, the transmission unit 35 of the recommendation device 30 transmits the transcription result and the property needs information 32c to the language processing device 60. Then, the instruction unit 36 ​​of the recommendation device 30 instructs the language processing device 60 to extract latent needs of the customer.

[0078] In step S119, the receiving unit 61 of the language processing device 60 receives the transcription result and the property needs information 32c. The AI ​​analysis unit 62 of the language processing device 60 analyzes the transcription result and updates the property needs information 32c by reflecting the customer's potential needs, values, lifestyle, and the like. Specifically, as shown in FIG. 9(b), the AI ​​analysis unit 62 reflects "Kansai," "Osaka," and "Moriguchi" as area information in the property needs information 32c based on the customer's comment, "An apartment in Moriguchi, Osaka." The AI ​​analysis unit 62 also sets the property category to "apartment" based on the customer's mention of the property category, "apartment." The AI ​​analysis unit 62 also determines, based on the customer's comment, "Apartment because I work from home a lot," as shown in FIG. 9(c), that the customer desires a quiet property and reflects this in the comment. Furthermore, as shown in FIG. 9(d), based on the customer's statement that "I want to listen to music in my room...", the AI ​​analysis unit 62 infers that the customer desires a reinforced concrete (RC) structure with excellent soundproofing, and reflects this in the structure. Then, as shown in FIG. 9(e), based on the customer's statement that "I have two naughty children...", the AI ​​analysis unit 62 determines that the customer is a family of three or more and requires a larger room, and sets "60m" as the lower limit of the desired area. 2 " In this way, the AI ​​analysis unit 62 gradually enriches the property needs information 32c while taking into account the customer's direct statements and contextual meaning. As a result, the customer's desired conditions, which were initially unclear, are dynamically supplemented and clarified as the conversation progresses. This allows for the construction of highly accurate property needs information 32c.

[0079] In step S120, the update unit 63 of the language processing device 60 updates the property needs information 32c based on the processing result of the AI ​​analysis unit 62.

[0080] In step S121, the transmission unit 64 of the language processing device 60 transmits the updated property needs information 32c to the recommendation device 30.

[0081] In step S122, the property needs information acquisition unit 41 of the recommendation device 30 receives the updated property needs information 32c transmitted from the language processing device 60 via the receiving unit 34. Then, the property needs information acquisition unit 41 of the recommendation device 30 updates the original property needs information 32c stored in the storage unit 32 by the updating unit 37 to the received property needs information 32c. This allows the property needs information 32c to be constructed in which the contextual conditions and wishes based on the customer's utterance are accurately reflected.

[0082] In step S123, the property extraction unit 42 of the recommendation device 30 performs collaborative filtering using the updated property needs information 32c. Specifically, in collaborative filtering, the property extraction unit 42 vectorizes the property needs information 32c of the target customer and the property needs information of multiple other customers stored in the storage unit 32, and calculates the cosine similarity (e.g., weighted correlation coefficient) between them to quantitatively evaluate the similarity of preferences with other customers. Note that each vector component used in calculating the cosine similarity is multiplied by the importance weight set by the setting unit 39. This realizes a weighted similarity evaluation that reflects the priority between items. For example, a large weight is assigned to items that the customer particularly values, such as "structure" and "area," and a small weight is assigned to items with lower priority, resulting in a structure in which high-weight items contribute more strongly to the similarity calculation. These weights determine the order of items in the property needs information 32c. That is, in FIG. 8, items appearing earlier are assigned higher weights.

[0083] In step S124, the property extraction unit 42 of the recommendation device 30 executes rule-based matching. For example, when the property needs information 32c is as shown in FIG. 9, the property extraction unit 42 extracts properties that are not reinforced concrete in structure or have an area of ​​60 m or less. 2 Properties that are less than this amount or are in a real estate category other than condominiums will be excluded from the recommended candidates.

[0084] In step S125, the property extraction unit 42 of the recommendation device 30 performs content-based filtering. In the content-based filtering, the property extraction unit 42 vectorizes the property needs information 32c and the multiple property information 32b stored in the storage unit 32, and evaluates directional proximity by calculating cosine similarity (e.g., weighted correlation coefficient) between the needs vector and each property vector. Note that each vector component used in calculating the cosine similarity is multiplied by the importance weight set by the setting unit 39. This vectorization employs a structure that allows multiple property candidates to be simultaneously evaluated for one property needs information 32c. Furthermore, distances and angles are calculated collectively in a weighted vector space multiplied by the importance weight for each item. This achieves highly accurate scoring that reflects the weighted compatibility between attributes.

[0085] In this way, the recommendation device 30 extracts recommendation results (a group of properties to be recommended to the customer) through the processes of steps S123 to S125.

[0086] In step S126, the output unit 44 of the recommendation device 30 outputs a pop-up display to the negotiation device 10.

[0087] In step S127, the display processing unit 21 of the negotiation device 10 displays a pop-up in the lower right corner of the screen of the display device 12, as shown in Fig. 11. This pop-up displays a notification such as "We have matching properties!", and is configured so that the salesperson can immediately grasp whether there are any recommended results even while talking with the customer. This allows for a smooth transition to property proposals that match the customer's interests and concerns, without interrupting the flow of negotiations.

[0088] In step S128, the transmitting unit 35 of the recommendation device 30 transmits the recommendation result to the language processing device 60.

[0089] In step S129, the instruction unit 36 ​​of the recommendation device 30 instructs the language processing device 60 to generate a reason for recommendation, as shown in FIG.

[0090] In step S130, the AI ​​analysis unit 62 of the language processing device 60 analyzes the relevance of the property information 32b extracted in steps S123 to S125 with the property needs information 32c updated in step S120, and generates a reason for suitability for each property. For example, as shown in Fig. 8, the AI ​​analysis unit 62 generates a textual recommendation reason such as "a property located in a quiet residential area is a good match for a customer who desired a quiet living environment" or "a property with an RC structure that has excellent soundproofing was presented in response to an utterance that suggests a preference for RC structures."

[0091] In step S131, the transmission unit 64 of the language processing device 60 transmits the reason for recommendation generated by the AI ​​analysis unit 62 to the recommendation device 30.

[0092] In step S132, the reason acquisition unit 43 of the recommendation device 30 receives the recommendation reason from the language processing device 60 via the reception unit 34. Then, the transmission unit 35 of the recommendation device 30 transmits the recommendation result extracted in the processes of steps S123 to S125 and the received recommendation reason to the negotiation device 10.

[0093] In step S133, the receiving unit 20 of the negotiation terminal 10 receives the recommendation results from the recommendation device 30. Then, the display processing unit 21 of the negotiation terminal 10 displays the recommendation results in the right half of the screen of the display device 12, as shown in FIG. 12. The recommendation results are displayed in a card-type display area in descending order of score, such as property A (score: 98), property B (score: 96), property C (score: 93), etc. The display processing unit 21 displays the property name or identification information, score value, and main features of the property (e.g., floor plan, location, structure, overview of the surrounding environment, etc.) in the display area.

[0094] The salesperson can explain the basis for the recommendation on the spot while showing the screen in sync with the customer's conversation. Additionally, to support the score, the reasons why the property matches the customer's property needs information 32c, such as "a quiet environment" or "a layout suitable for raising children," are briefly displayed based on the analysis results of the customer's latent needs extracted by the language processing device 60 and the customer's emotional tendencies extracted by the voice recognition device 50. This makes it easier for the customer to understand the appropriateness of the proposed property on the spot, contributing to improving the accuracy and speed of sales negotiations.

[0095] In step S134, the transmission unit 35 of the recommendation device 30 transmits the reason for recommendation to the voice recognition device 50. Then, the instruction unit 36 ​​of the recommendation device 30 instructs the voice recognition device 50 to extract the customer's emotion and the intensity of the conversation.

[0096] In step S135, the emotion change extraction unit 54 of the voice recognition device 50 estimates a change in customer emotion by analyzing the tone of each speaker's speech, tone of speech, pauses, the presence or absence of exclamations, intonation information at the end of sentences, etc., based on the conversational voice at the time of transcription. An emotion change is defined as a change from a positive emotion to a negative emotion, a change from a negative emotion to a positive emotion, a change from an interested emotion to an uninterested emotion, a change from an uninterested emotion to an interested emotion, etc.

[0097] In step S136, the energy extraction unit 55 of the voice recognition device 50 calculates the customer's speech energy based on the conversational voice. Specifically, the energy extraction unit 55 comprehensively evaluates the frequency of intensifiers (such as "very," "quite," and "really") and positive adjectives (such as "nice," "great," and "sounds good") contained in the sentence spoken by the customer, the volume and intonation of the voice, changes in speech speed, and the extension of word endings, and generates a scored energy index. Furthermore, factors such as repeated speech on the same topic and successive high-pitched sounds appearing in the speaker's speech waveform are also taken into account as factors that determine whether speech has high energy. The energy calculated in this way is quantitatively expressed as a numerical value (for example, in the range of 0 to 100).

[0098] In step S137, the transmission unit 56 of the voice recognition device 50 transmits the extracted emotion and enthusiasm of the customer to the recommendation device 30.

[0099] In step S138, the energy acquisition unit 47 of the recommendation device 30 receives the emotional change and the energy index of the customer from the voice recognition device 50 via the receiving unit 34. Then, the determination unit 48 of the recommendation device 30 determines whether the energy index is equal to or greater than a predetermined threshold value.

[0100] If the determination unit 48 makes a negative determination, the recommendation device 30 returns. On the other hand, if the determination unit 48 makes a positive determination, in step S139, the information receiving unit 46 of the recommendation device 30 receives the change in the customer's emotion via the receiving unit 34. Then, the determination unit 48 determines whether the customer's emotion has improved.

[0101] If the determination unit 48 makes a negative determination, the recommendation device 30 returns. On the other hand, if the determination unit 48 makes a positive determination, in step S140, the output unit 44 of the recommendation device 30 outputs an instruction to update the recommendation result and the reason for recommendation to the negotiation device 10.

[0102] In step S141, the display processing unit 21 of the negotiation terminal 10 updates the display of the recommendation result and the reason for recommendation.

[0103] In step S142, the sending unit 35 of the recommendation device 30 sends the extracted result of the customer's emotional change to the language processing device 60. Then, the instructing unit 36 ​​of the recommendation device 30 instructs the language processing device 60 to create the text of the proposal email.

[0104] In step S143, the AI ​​analysis unit 62 of the language processing device 60 generates the text of a proposal email to the customer based on the recommendation results and reasons for recommendation received from the recommendation device 30. The AI ​​analysis unit 62 automatically generates text that describes information about properties that meet the customer's desired conditions (such as property name, location, structure, layout, and price) in a natural context while citing the reasons for recommendation generated by itself. In addition, the AI ​​analysis unit 62 references the output result of the emotion change extraction unit 54 received from the recommendation device 30 along with instructions for creating the text of the proposal email, thereby generating text that includes expressions that more actively appeal to properties to which the customer responded favorably (e.g., "This is a property in the XX area that you were particularly interested in")

[0105] In step S144 , the sending unit 64 of the language processing device 60 sends the text of the proposal email to the recommendation device 30 .

[0106] In step S145, the email text acquisition unit 45 of the recommended device 30 receives the text of the proposed email from the language processing device 60 via the receiving unit 34. Then, the sending unit 35 of the recommended device 30 sends the proposed email to the email address of the sales representative included in the sales representative information 32a.

[0107] The information processing system 1 transcribes customer conversations in real time and applies the transcription results to AI analysis means, thereby extracting not only explicit desired conditions but also latent needs from context and speech tendencies. As a result, it becomes possible to recommend properties that match preferences and lifestyles that the customer himself is not aware of, realizing highly accurate needs understanding and immediate proposals that were difficult to achieve with conventional static search methods.

[0108] Furthermore, according to this embodiment, output data including the recommendation reasons generated by the language processing device 60 is sent to the user terminal via the output processing unit, allowing the salesperson to instantly present recommendation reasons that match the customer's preferences and wishes. This clarifies the basis for the proposal, allowing the salesperson to provide a convincing explanation to the customer and maintaining a high level of consistency and quality in the sales pitch. Furthermore, the explanation burden on the salesperson is reduced, thereby improving the efficiency of the proposal work.

[0109] Furthermore, according to this embodiment, the recommendation device 30 outputs an instruction to create property needs information 32c to the language processing device 60 along with the transcription result, and receives the analysis result. This allows the transcription process and the needs extraction process to be linked with high efficiency. As a result, customer needs information is constructed stepwise and dynamically in line with the progress of the conversation, achieving highly accurate matching without sacrificing real-time performance.

[0110] The language processing unit 60 then extracts multiple attributes from the transcription results, such as desired area, property type, structure type, price range, yield, floor area, and additional requirements, and generates property needs information as structured data. This converts the customer's speech into comprehensive and formally organized data. This improves the accuracy of condition matching in the matching processing unit, enabling advanced recommendations that meet the customer's needs. Furthermore, the structured needs information can be used for comparative analysis with other customers and for future statistical processing, making it a highly versatile information asset.

[0111] Furthermore, the recommendation device 30 combines collaborative filtering, rule-based matching, and content-based filtering to make optimal property recommendations that comprehensively consider statistical trends, explicit conditions, and contextual features, thereby enabling flexible and accurate matching even for complex needs that are difficult to address using individual methods.

[0112] Furthermore, the recommendation device 30 re-executes matching each time property needs information 32c is registered, and if a match is established by comparing it with existing customers and properties, it automatically generates the text of a proposal email and notifies the customer. This significantly reduces the burden of information search and follow-up work on sales representatives, while also creating timely proposal opportunities and continuously improving proposal accuracy.

[0113] Furthermore, in the speech recognition process, the speech recognition device 50 performs segmentation based on silence detection and sentence completeness rather than simple time-based segmentation, enabling transcription at semantic units. As a result, the accuracy of AI analysis is improved, and the reliability of extracted keywords and emotions is increased.

[0114] Furthermore, the language processing device 60 not only receives transcribed conversation information as input, but also receives sentiment analysis results. This enables dynamic matching that reflects the tone and intent of the entire conversation. This dramatically improves the accuracy and persuasiveness of the recommendation results.

[0115] Furthermore, the negotiation device 10 dynamically controls the timing of transmission to the voice recognition device according to the size of the voice data and the processing load. Furthermore, the voice recognition device 50 detects silence for a certain period of time (for example, 5 seconds) or more and determines the end of a sentence, such as punctuation, to process the voice by dividing it into meaningful speech units, organize the speech content by speaker, and transmit the organized speech content by speaker to the recommendation device 30. This dynamically adjusts the timing of transmission of the transcription results from the recommendation device 30 to the language processing device 60. This achieves distribution of the processing load of the entire system, improves responsiveness, and enables stable operation without impairing real-time performance.

[0116] It goes without saying that the present invention is not limited to the above-described embodiment, and can be embodied in various forms as long as they fall within the technical scope of the present invention.

[0117] For example, in the above-described embodiment, the recommendation device 30 extracts the recommendation results by performing rule-based matching, content-based filtering, and collaborative filtering. However, the recommendation device 30 may extract the recommendation results by performing two of these. Furthermore, these processes may be performed by the language processing device 60.

[0118] In the above-described embodiment, the recommendation device 30 transmits the recommendation result and the display update to the negotiation device 10 when the customer's emotions improve. However, the recommendation device 30 may transmit the recommendation result and the display update instruction to the negotiation device 10 at other times. For example, the recommendation device 30 may transmit the recommendation result and the display update instruction to the negotiation device 10 every time the property needs information 32c received from the language processing device 60 is updated. Alternatively, the recommendation device 30 may transmit the recommendation result and the display update instruction to the negotiation device 10 at the time when an update instruction is input from the operation device 14 connected to the negotiation device 10.

[0119] In the above-described embodiment, the negotiation device 10 is shared by both the sales representative and the customer. However, the negotiation device 10 may be divided into a terminal for the sales representative and a terminal for the customer. In this case, the speech recognition device 50 can more easily perform speaker separation.

[0120] In the above-described embodiment, when the property information 32b is updated, the recommendation results for existing customers may be updated, and the updated recommendation results may be transmitted to the customers.

[0121] In the above-described embodiment, the recommendation device 30 may send a proposal email not only to the sales representative but also to the customer.

[0122] In the above-described embodiment, the recommendation device 30 instructs the speech recognition device 50 to extract changes in the customer's emotions. However, the recommendation device 30 may instruct the language processing device 60 to extract changes in the customer's emotions. In this case, the language processing device 60 estimates the customer's emotional tendency by analyzing the context of the utterance, tone of voice, vocabulary selection, the presence or absence of exclamations or negative expressions, etc. For example, when a positive expression such as "That's nice, I'm interested in that" or a negative or ambiguous expression such as "Hmm, a little..." appears, the language processing device 60 infers that a change in emotions has occurred, triggered by the appearance of such an expression.

[0123] In the above-described embodiment, the recommendation device 30 instructs the speech recognition device 50 to analyze emotions, receives emotional information, and uses it to update the display of recommendation results. However, the recommendation device 30 may instruct the speech recognition device 50 or the language processing device 60 to extract keywords based on the content of a conversation with a customer and receive the extracted results. In this case, visual presentation that matches the customer's interests, such as preferentially displaying property information related to the extracted keywords or highlighting the card area of ​​the property, is possible, thereby improving the accuracy of the salesperson's explanation and the persuasiveness of the proposal. Furthermore, the recommendation device 30 may instruct the speech recognition device 50 or the language processing device 60 to extract topic information from the conversation and receive the extracted topic information. In this case, recommended properties can be grouped and displayed by category according to the topic classification, and the display order of each group can be optimized based on the customer's speech tendencies and preferences. This enables information presentation that is in line with the customer's cognitive characteristics and consideration process, further enhancing the persuasiveness of proposals. The recommendation device 30 may also instruct the speech recognition device 50 or the language processing device 60 to perform two or more of sentiment analysis, keyword extraction, and topic extraction, and receive the results of a composite analysis. This allows, for example, properties that match keywords or topics included in utterances in which the customer expressed favorable feelings to be displayed with a higher priority, or to be presented with a label indicating a favorable emotional response (e.g., "high interest"). This configuration makes it possible to provide an interface that is in tune with the customer's emotions and context, thereby providing strong support for decision-making during business negotiations.

[0124] In the above-described embodiment, one piece of property needs information 32c is created for one customer. However, multiple pieces of property needs information 32c may be created for one customer. In this case, the recommendation device 30 may extract one recommendation result across the multiple pieces of property needs information 32c, or may extract a recommendation result for each of the multiple pieces of property needs information 32c.

[0125] In the above-described embodiment, the present invention has been described as an information processing system. However, the present invention may also be an information processing method executed by the information processing system. Alternatively, the present invention may also be a program for executing the information processing method.

[0126] In the above-described embodiment, the negotiation terminal 10 is described as a general-purpose computer. However, the negotiation terminal 10 may be a tablet terminal, a smartphone, or the like.

[0127] In the above-described embodiment, the AI ​​analysis unit (AI analysis) has been described as performing various processes using a large-scale language model. However, the AI ​​analysis unit (AI analysis) may perform processes using, for example, a machine learning model, a deep learning model, or other models, and each process may be implemented using a single or multiple neural networks or a combination of individual algorithms. Furthermore, the AI ​​analysis may use these models to perform, for example, the following processes: AI analysis may perform preprocessing / feature extraction to extract acoustic features (MFCC (Mel-Frequency Cepstral Coefficient), pitch, etc.) and linguistic features (BERT Embedding, etc.) from audio waveform or character string data. The AI ​​analysis may perform intent estimation / keyword extraction processing to estimate the intent or keywords of customer utterances using the extracted features as input. The AI ​​analysis may perform emotion and energy estimation processing to estimate emotion categories and energy scores from speech text and acoustic features. The AI ​​analysis may perform a needs generation process to generate or update each item of property needs information based on estimated keywords, intentions, etc. The AI ​​analysis may evaluate the relevance between the extracted property candidates and the needs information as a recommendation reason generation process, and generate the reason in natural language. Each of the above processes may be implemented using a single or multiple neural networks, or may be implemented by combining individual algorithms. [Explanation of symbols]

[0128] 1 Information processing system, 10 Business negotiation device, 11 Communication unit, 12 Display device, 13 Voice input device, 14 Operation device, 15 Memory unit, 16 Control unit, 17 Reception unit, 18 Voice storage processing unit, 19 Transmission unit, 20 Reception unit, 21 Display processing unit, 22 Transcription result display unit, 30 Recommendation device, 31 Communication unit, 32 Memory unit, 32a Sales representative information, 32b Property information, 32c Property needs information, 32d Property information, 33 Generation unit, 34 Reception unit, 35 Transmission unit, 36 Instruction unit, 37 Update unit, 38 Storage processing unit, 39 Setting unit, 40 Transcription result acquisition unit, 41 Property needs information acquisition unit, 42 Property extraction unit, 43 Reason acquisition unit, 44 Output unit, 45 Email text acquisition unit, 46 Information reception unit, 47 Energy acquisition unit, 48 Judgment unit, 50 speech recognition device, 51 receiving unit, 52 reflection unit, 53 transcription processing unit, 54 emotion change extraction unit, 55 heat extraction unit, 56 transmitting unit, 60 language processing unit, 61 receiving unit, 62 AI analysis unit, 63 update unit, 64 transmitting unit, N network.

Claims

1. An information processing system for matching real estate properties, a storage unit for storing customer property needs information and property information; a transcription result acquisition unit that acquires real-time transcription results of conversational voices with customers; a property needs information acquisition unit that acquires property needs information of the customer automatically extracted by AI analysis based on the transcribed conversation content; a property extraction unit that extracts properties that match the property needs information of the customer from the property information stored in the storage unit based on the property needs of the customer; a reason acquisition unit that acquires a recommendation reason generated by the AI ​​analysis based on the characteristics of the extracted property and the property needs information of the customer; an output unit that outputs the recommendation result including the recommendation reason to a user terminal; Equipped with the property extraction unit analyzes the compatibility between the property information and the customer's property needs information using at least two or more methods of rule-based matching, content-based filtering, and collaborative filtering, and combines the results of these analyses to extract the properties; the property extraction unit, in the collaborative filtering, extracts the property based on a correlation coefficient between the vectorized property needs information of the customer and the property information, and extracts the property based on a correlation coefficient between the vectorized property needs information of the customer and property needs information of other customers; the property extraction unit calculates similarities in the content-based filtering and the collaborative filtering based on importance weights preset for each of a plurality of items included in the property needs information of the customer; Information processing system.

2. 2. The information processing system according to claim 1, The property needs information acquisition unit instructs the AI ​​analysis to generate property needs information of the customer together with the transcription result, and acquires the analysis result by the AI ​​analysis as the property needs information of the customer. Information processing system.

3. 3. The information processing system according to claim 1, The customer's property needs information is generated by converting items including the desired area, category, structure, price range, yield, area, and supplementary requirements from the transcription results into structured data using the AI ​​analysis. Information processing system.

4. 3. The information processing system according to claim 1, the user terminal has a display unit capable of displaying various types of information; The output unit causes the display unit to display the transcription result and the recommendation result. Information processing system.

5. 2. The information processing system according to claim 1, a setting unit for setting the importance weight in accordance with an operation by an operator; An information processing system comprising:

6. 3. The information processing system according to claim 1, the output unit outputs a notification to the user terminal together with the recommendation result when the property extraction unit extracts a property that matches the property needs information of the customer. Information processing system.

7. 3. The information processing system according to claim 1, The storage unit, When new property information is registered and the new property information is extracted by the property extraction unit, or When property needs information of a new customer is registered and a property is extracted by the property extraction unit When at least one of the above conditions is met, a proposal email containing the recommendation result is obtained. Email text acquisition section An information processing system comprising:

8. 2. The information processing system according to claim 1, The transcription results are: When a silent section that continues for a predetermined time or more is detected in the audio data, or If the end of at least one sentence is detected in the audio data When any of the following conditions is met, the utterance is divided into units of speech. Information processing system.

9. 9. The information processing system according to claim 8, A setting unit for setting the delimiting conditions in accordance with an operation by an operator. An information processing system comprising:

10. 2. The information processing system according to claim 1, an information receiving unit that, when the property needs information of the customer is updated by the AI ​​analysis, additionally receives, in addition to the transcription result, at least one of keywords extracted from the conversational voice, speaker emotion information, and conversation topic information; an energy acquisition unit that acquires the energy of the customer estimated based on the keyword, the emotion information, or the topic information; a determination unit that determines whether the amount of heat exceeds a predetermined threshold; Equipped with the output unit instructs the user terminal to update the recommendation result when the determination unit makes a positive determination. Information processing system.

11. An information processing method for matching real estate properties, the method being executed by an information processing system, a storage step of storing customer property needs information and property information; a transcription result acquisition step of acquiring information transcribed in real time from a conversation voice with a customer; a property needs information acquisition step of automatically extracting property needs information of the customer from the transcribed conversation content by AI analysis and acquiring the property needs information of the customer; a property extraction step of extracting properties that meet the property needs of the customer from the property information stored in the storage step based on the property needs of the customer; a reason acquisition step of acquiring a recommendation reason automatically generated by the AI ​​analysis for the extracted property based on the property's characteristics and the customer's property needs; an output step of outputting the recommendation result including the recommendation reason to a user terminal; Equipped with The property extraction step analyzes the compatibility between the property information and the customer's property needs information using at least two or more methods of rule-based matching, content-based filtering, and collaborative filtering, and combines the results of these analyses to extract the properties; The property extraction step extracts the property based on a correlation coefficient between the vectorized property needs information of the customer and the property information in the collaborative filtering, and extracts the property based on a correlation coefficient between the vectorized property needs information of the customer and property needs information of other customers; the property extraction step calculates similarities in the content-based filtering and the collaborative filtering based on a predetermined importance weight for each of a plurality of items included in the property needs information of the customer; Information processing methods.

12. A program for causing a computer to execute each step of the information processing method according to claim 11.

Images