JP2021131908A - Natural language grammars adapted for interactive experiences - Google Patents

Abstract

To provide an advertiser with an advantage by accurately controlling the time of distribution of an advertisement is distributed so that the advertisement is more effective.SOLUTION: Natural language grammars interpret expressions at conversational human-machine interfaces of devices. Under conditions favoring engagement as specified in a unit of conversational code, the device initiates a discussion using one or more of TTS, image, video, audio, and animation depending on device capabilities of a screen and audio output. Conversational code units specify conditions on the basis of conversation state, mood, and privacy. Grammars provide intents that cause calls to system functions. Units can provide scripts for guiding the conversation. The device or supporting server system can provide feedback to creators of the conversational code units for analysis and machine learning.SELECTED DRAWING: Figure 2

Description

Field of Invention The present invention relates to the field of speech-based conversational interfaces between humans and machines using natural language grammar. (Category 704 / E17.015 or 704/251)

Background David Sarnoff, an early pioneer in radio and television broadcasting technology, said before becoming the president of RCA® and the founder of NBC®, "Radio players can imagine products. Not worth it. Does anyone pay for a message that hasn't been addressed to someone specific? " He was still unaware of the upcoming giant industry of sending advertisements through telecommunications and networks.

Visual advertising was included in printed publications for about the same period as the publications began to be printed. It must have been a surprise to Mr. Sanov that it became commonplace to broadcast audio advertisements several times an hour in other radio broadcast content in the end. Similarly, television broadcasters have adopted a method of broadcasting video advertisements several times an hour within television broadcast content. Naturally, with the advent of internet-connected computers, mobile phones, tablet devices, music players, virtual reality, and other such devices, visual and video advertising studded with other content desired by device users. , And voice ads were served through the device.

Whether in print media, radio, television, or devices connected to the Internet, advertising is desirable, accepting the highest bids, advertisers who bid on advertising opportunities by offering advertising and money. Publishers that deliver related ads between content to potential customers (lead generation) of the advertiser's products or services, and hopefully convert the potential customers to customers who will make money for the advertiser. It consists of things.

In print, radio, and television ads, there was no direct way to identify how many potential customers were involved in an ad by reading or focusing on broadcast messages. There was no direct way to characterize engagement, such as when and where engagement occurred, and the type of potential customer. Internet-based devices have provided the ability to directly identify and characterize potential customer engagement. That is, it is generally the act of a potential customer clicking on a link in a browser. As information requests are sent to advertisers by clicking on links, the Internet has provided a way to measure engagement and more accurately predict conversions (purchases of products or services advertised).

Another valuable form of advertising is product placement. That is, the product or service in the advertisement is included as part of the story that is the content. For example, the movie E.I. T. Then, extraterrestrials chased the traces of candy of the brand Lacy's Pieces (registered trademark), and in the TV drama Sex and the City, the character Carrie frequently showed shoes of the brand Manolo Blahnik. I was talking about it.

Another form of advertising is telemarketing. That is, to call a potential customer. Telemarketing has a reputation for being annoying to potential customers. Because, at least, it's not at the right time for potential customers. In addition, it is a difficult selling point because it occurs in the form of one uninterrupted conversation. Telemarketing, as we all know, is ineffective, has low conversion rates, requires relatively expensive human operators to handle unpredictable reactions of potential customers, and is a relatively simple manuscript that human operators can follow. It requires, wears down the patience and psychological health of the operator, and is unable to reach the most profitable potential customers.

Invention Overview Currently, most advertisements are not interactive. There are still image ads, voice message ads, and video ads. Considerable research has been done on accurately targeting potential customers with non-interactive advertising. Most of the interactivity in conversational ads is the ability to click one or a few of the buttons in a visual ad to see a small amount of specific additional information or to play a game. ..

This disclosure defines ad units that support interaction with potential customers in a conversational natural language, serves ads, bids on ad inventory, sends information from potential customers to publishers, and from publishers to advertisers. It covers systems and methods that provide, machine-learn from customer interactions, and are relevant to the advertising ecosystem.

By using certain embodiments, ad conversions can be attributed accurately and meaningfully to the acquisition path. The higher the conversion rate and the more measurable, the more influential and accurate the targeting will be.

Delivering interactive ads in natural language from a trusted machine interface is the final form of product placement in which the posted content becomes the behavior of the machine and the product is placed surrounded by the content.

It is a figure which shows the engagement between the conversational robot apparatus and a potential customer which concerns on embodiment. It is a figure which shows the flowchart of the method of providing the experience which concerns on embodiment. It is a figure which shows the dialogue between the human and the machine with a conversation state, and the bidding based on the conversation state which concerns on embodiment. It is a figure which shows the interpretation of the expression with respect to the background of the conversation state which is also used for analyzing the bid which concerns on embodiment. It is a figure which shows the dialogue script which concerns on embodiment. FIG. 5 is a diagram showing a state in which an advertiser bids to deliver an advertiser's advertisement to a potential customer according to the embodiment. It is a figure which shows the flow of the advertisement from the advertiser to the potential customer through the advertisement ecosystem which concerns on embodiment. FIG. 5 illustrates the delivery of interactive advertising with action and conversational feedback through an ecosystem, according to an embodiment. It is a figure which shows the system for the exchange between a potential customer and an advertisement server which concerns on embodiment. It is a figure which shows the non-temporary computer readable medium which is a rotating disk which concerns on embodiment. It is a figure which shows the non-temporary computer readable medium which is a flash random access memory which concerns on embodiment. FIG. 5 shows the underside of a computer processor-based system-on-chip according to an embodiment. It is a figure which shows the upper side of the computer processor-based system-on-chip which concerns on embodiment. It is a figure which shows the server which concerns on embodiment. It is a block diagram of the system-on-chip for a device which concerns on embodiment. It is a block diagram of the server processor which concerns on embodiment. It is a figure which shows the component of the conversational robot apparatus which concerns on embodiment. It is a figure which shows the voice assistant apparatus which concerns on embodiment. It is a figure which shows the voice earphone device which concerns on embodiment. It is a figure which shows the mobile phone apparatus which concerns on embodiment. It is a figure which shows the automobile device which concerns on embodiment.

Detailed Description According to various embodiments, a natural language expression is a human being for information on one or more tokens by speaking or by other means such as typing, tapping, gestures, or direct cognitive coupling. Can be an expression of. Natural language grammar can be one or more intents, each of which has a set of possible expressions associated with the intent. The field can be a grammar set specific to a conversation topic or dataset. An intent can be a data structure that represents a human hypothesized desired action represented by a natural language representation. Some examples of actions include retrieving information and responding to inquiries, sending messages, performing actions, or SDKs for Web APIs (Application Programming Interfaces) to perform any desired action. Software Development Kit) function call may be made. Interpretation can be the process of deriving intents from token sequences or token sequence hypotheses.

According to some embodiments, the advertiser is the entity that defines the ad unit and pays for the ad unit to be delivered to potential customers. An ad unit consists of content, bids, and one or more natural language grammars. According to some embodiments, the ad begins with delivering a referral message defined as content within an interactive ad unit. The introductory message is content intended to initiate a concrete human-machine interaction. Some introductory messages are as short as one phrase. Content can be one or more of images, videos, audio clips, spoken language, etc., according to various embodiments, and content is meaningful to potential customers. Some introductory messages are as long as conversational commercial messages. Some embodiments present an introductory message to initiate a new conversation. Some embodiments turn the introductory message into a conversation as part of the machine's conversational turn, such as responding to a request for an ingredient in a cooking recipe with a specific brand name for the ingredient.

According to some embodiments, the publisher is the entity that delivers the advertisement to the potential customer and is rewarded for doing so. In general, publishers serve such advertisements among other content that potential customers want, expect, or demand. In some embodiments, the publisher utilizes a server to control or supply content destined for one or more remote devices. In some embodiments, the device performs the publisher function on its own. In some embodiments, the publisher works with one or more intermediaries to determine which ads to accept or deliver to potential customers. In some embodiments, the publisher accepts information from potential customers or users of devices that deliver advertisements, or accepts information about potential customers or users of devices that deliver advertisements. Some examples of device embodiments include humanoid robots, home appliances, household assistant devices, wearable electronic devices, mobile phones, desktop computers, automobiles, outdoor signage, kiosks, vending machines, human-machine interfaces. There are devices and so on.

According to some embodiments, after the referral message has been delivered, or after the referral message has begun to be delivered, the publisher identifies a natural language representation of engagement provided by the potential customer. Such an embodiment interprets a natural language expression according to a natural language grammar defined within an interactive ad unit. In response to the interpretation that the natural language expression is consistent with the grammar, such an embodiment can be determined to have engaged.

An example of an expression that some embodiments can handle is an engagement question that triggers a conversation, such as:

"Tell me more [about that ad | product]"
"What was that [ad]".

Personal assistant engagements such as:
"Email me a link"
"Remind me about this [tonight, tomorrow]".

Engagement of general retail questions such as:
"Show [it to] me"
"Where can I buy it"
"[How much | what] does it cost ([how much | what] does it cost)"
"Buy it [online] [now]".

Product-specific engagement questions such as:
"How many rolls is that"
"What colors does it come in"
"How much does it weigh".

Or another engagement question for which no response is available.
Some embodiments utilize an intermediary. Brokers connect advertisers to publishers, collect ad units, collect inventory, analyze bids based on lead attributes, context, conversational state, mood, etc., depending on supply and bids. It facilitates delivery of ad units to inventory by setting prices, providing advertisers with information about potential customer engagement and conversions, and more. Engagement methods differ between embodiments. According to an embodiment using a voice-enabled conversational device, engagement is a read-aloud response interpreted by the ad unit grammar.

Various embodiments utilize different modes and means of communication in natural language. Some of them are verbal, using TTS (Text-To-Speech) for the machine to output speech and automatic speech recognition (ASR) for the machine to recognize human speech. Such an embodiment provides an audio experience. Some embodiments utilize music, sound effects, or other audio information to provide an audio experience. Some embodiments utilize text input / output modes, including text display and transmission, for interaction. Some embodiments provide natural language-based human-machine interaction for other neural sensing and neurostimulation modes.

Some embodiments utilize a display device in place of or in addition to an audible machine output means such as a speaker and in place of the ASR of an audio signal captured from a microphone or other barometric pressure wave sensing device. In addition, there are also embodiments that utilize text input means such as a keyboard or touch screen.

Ads that are read aloud by the machine will be more compelling if they are delivered in the voice that potential customers normally expect from the machine. Some embodiments that serve ads using TTS to generate speech output are sufficient by accepting text with metadata, such as markup languages such as Speech Synthesis Language (SSML). conduct. This allows you to set a tone that is useful for attracting and persuading potential customers, or emphasizing specific words, even though you are using the voice of the machine.

Many find it easier to specify their own intent by continuously getting the information they need to be fully aware of the intent. Modal dialogs can facilitate such interactions. A modal dialog can be a sequence of human-machine interactions that alternates the machine with the user and prompts the user to provide information or to perform the action someone desires. For example, to trigger the intent of sending a text message, the machine would sequentially ask the user to provide the recipient's name and then send the message. As another example, to calculate monthly mortgage payments, the machine will sequentially ask users to provide home value, down payment percentages, interest rates, loan terms, and other information. Some users may be frustrated when another interaction interrupts a modal dialog. When the conversational interface is not involved in the modal dialog, some embodiments only deliver an introductory message. In some embodiments, when they notice an opportunity to deliver an introductory message, they deliver it as soon as the modal dialog is complete.

Some embodiments may interrupt the original activity, initiate engagement, and request a response before continuing the original activity, even outside the modal dialog. For example, while using a machine capable of playing a game, some embodiments interrupt the game and deliver an introductory message casually. Some such embodiments have a timeout, after which the original activity resumes in the absence of potential customer engagement.

Various embodiments provide an advantage to the advertiser by precisely controlling when the advertisement is delivered so that the advertisement is more effective. Some embodiments deliver some advertisements not immediately, but when the conversation shows usefulness. For example, if the weather is raining, a conversation with the machine about the weather will cause the machine to serve an umbrella ad only when the potential customer is preparing to go out.

According to some embodiments, there are multiple conditions for ending the advertising conversation. A timeout at the end of a period of no voice activity can indicate ending an advertising conversation. According to some embodiments, the timeout period is 5 seconds, but shorter or longer periods are appropriate for different embodiments. In some embodiments, if the machine's turn in the advertising conversation ends with a question, the timeout is longer than if the machine's turn ends with an opinion. Some embodiments require and provide a response to potential customers to use wake expressions such as "ok robots" to avoid accidentally voice-recognizing inaudible words in the background. In some embodiments, a button click on the screen can be used to initiate a response.

According to some embodiments, the display of lack of interest ends the advertising conversation. For example, a lead expression with a high score in field grammar rather than advertising grammar may indicate a lack of interest in advertising. In some embodiments, the conversational state stores that the advertising conversation is in progress and, accordingly, gives greater weight to the advertising grammar when calculating the score compared to other discipline grammars. give.

According to some embodiments, the conversion ends the conversation. The ad unit may specify what constitutes a conversion. For example, a request in the navigation field for a store that sells an item can be a conversion, a request to order a product or service online can be a conversion, and a request for a later reply can be a conversion. .. In some embodiments, the ad unit maintains conversational engagement, such as dynamic small talk, indefinitely if the potential customer responds appropriately. In some such embodiments, the potential customer ends this conversation by timing out the system or by expressing that the score is high enough in areas other than advertising grammar.

Some embodiments consider ad units as an area. Many ad units are presented simultaneously to the natural language interpreter and compete with each other for the highest interpretation scores for each of the non-advertising areas and the lead's expression. Some such embodiments give the grammatical expression a dynamic weight for different possible wording about the intent.

Some embodiments utilize a grammar-based interpreter. Some embodiments make full use of a statistical language model-based interpreter.

Clickable ads contained in web pages were more effective than printed or television ads for visual communication methods, but the advantages of some embodiments of conversational ads are for the visually impaired and for the visually impaired. Improves the effectiveness of audio advertising for people who are visually involved in other things, such as driving a car or piloting a helicopter or space drone, at least as much as printed or television advertising. That is.

FIG. 1 is a diagram showing an engagement scenario. The humanoid robot-assisted natural language conversation posting device 11 with an advertising unit uses the built-in TTS software to send an introductory message saying "did you hear about the new StarPaste?" offer. The human potential customer 12 is involved by listening to this introductory message and saying "tell me about it".

FIG. 2 is a diagram showing a flowchart of an operation for delivering an interactive advertising experience. The system comprises a database 21 of ad units, each of which has a bid, content, and grammar. The process begins in step 22 with evaluating bids for ad units contained in database 21 and selecting the ad unit with the highest bid. In some embodiments, bid evaluation occurs offline long before the ad unit is delivered. In some embodiments, bid evaluation occurs in real time. The process proceeds to step 23 to deliver an introductory message for the ad unit selected for the highest bid. In step 24, after delivering the referral message, the system receives a natural language expression from the potential customer. The system proceeds to step 25 and interprets the natural language expression according to grammar 26 from the ad unit selected by the highest bid. If this expression matches grammar 26, the process proceeds to step 27. In step 27, the system performs the action indicated by the intent of grammar 26 with which the expression matches.

Some embodiments use multidimensional complex functions to determine in real time whether and when to deliver an advertising experience. In some embodiments, this function relates to how much the advertiser is willing to bid on the delivery of the ad, to estimate the likelihood that the ad will achieve conversions, and to use the posting device. Conversely concerns the estimation of the likelihood that an ad will discourage its use. These can be inputs to a choice function such as the Vickrey-Clarke-Gloves function. In various embodiments, the algorithm may be run periodically at the turn of each dialogue, or at the beginning and end of the conversation.

Estimating the likelihood of conversion can be a weighted function with various coefficients. Some factors relate to user interest in topics such as old conversation history, user preferences, user profiles (age, gender, ethnicity, purchase history, location), and history of engagement or refusal to similar ads. Some factors relate to recent conversations, such as recent conversation status history or keywords from recently accessed areas.

Estimating the likelihood of an ad that discourages users from using the placement device can be a weighted function with various coefficients. Some factors are the type of interrupt, such as whether there was an interrupt during no conversation, whether there was an interrupt at the assumed end of the conversation, or whether there was an interrupt in the middle of the conversation, and the conversation is trivial. Whether (for example, "good morning"), whether the conversation is informative (for example, "what's the weather and when is my next meeting?") When?) ”), Or whether it is an active conversation (for example,“ call my friend ”), regarding the type of conversation. Some factors relate to the history of user responses to ads, such as "shut up". Some factors are the length of the ad, whether it is an ad that requires a response to be sent, whether it is an interruptable ad, and the type of element it contains (eg, read-aloud text, music, video, etc.). Etc., regarding the degree of imposition of advertisements.

Conversational ads are most effective when they are most relevant to what the potential customer thinks when the ad is served. No ad is more appropriate than being served when its content matches the direct context of a potential customer or the direct topic of a conversation. Advertising is the best indicator available today to show what your potential customers are thinking when and when.

In some embodiments, the referral message of the advertisement includes content conditioned by context such as time, location, user profile, and conversational state. For example, an introductory message may begin by saying "good morning,""goodafternoon," or "good evening," depending on the time of day. The introductory message is "did you know that they have the new Star brand toothpaste at Wal-Mart?"
Did you know that there is a new ar brand toothpaste? ) ”Or“ did you know that they have the new Star brand toothpaste at Carrefour? (Carrefour?)
Did you know that ur® has a new Star brand toothpaste? ) ”, Etc., you may change the name of the store depending on the location and say the name of the nearest store that offers the product. The referral message may say the name of the potential customer based on the user profile. For example, "hey, Maria, did you hear about the new Star brand toothpaste?" Or "hey, Mario, did you hear about the new Star brand toothpaste?" (Hello, did you hear about Mario, the new Star brand toothpaste?) ". The introductory message says, "Before you go to the next Eagles game you should try the new Star brand toothpaste. "Before you go to tomorrow's party, you should try the new Star" or when you buy new clothes
You may mention specific entities included in recent conversational states, such as "brand toothpaste (try Star brand new toothpaste before going to tomorrow's party)".

Some embodiments that support conditional referral message content can specify conditions based on scope. For example, if the weather forecast says the temperature is between 20 ° C and 30 ° C (68 ° F and 86 ° F), the message will be "check out the new Star Wear styles of activewear". If the message is "check out the new Star Wear swimsuit lineup" or if the temperature forecast is higher than 30 degrees Fahrenheit (86 degrees Fahrenheit), check out the new Star Wear swimsuit lineup. If it is below 20 ° C (68 ° F), the message is "check out the new Star Wear down parka collection". ".

Some embodiments support conditional ad bid values based on conversational state. Some embodiments support threshold-based conditioning and others range-based conditioning. Some embodiments support conditional ad bid values based on non-conversational contexts.

In some embodiments, the conversational state is a data structure that stores variables and values related to the grammar. In some embodiments, the conversation state variable is deleted after a certain number of conversational dialogue turns. Some embodiments include a time stamp in the conversation state and remove the conversation state variable after a certain amount of time. Some embodiments may utilize different numbers or times of dialogue turns, depending on the type of variable or its value.

Some embodiments include one or more emotional values associated with various expressions stored in the conversational state. Some embodiments hold a current mood variable that represents the mood in an environment of human-machine interface.

In some embodiments, the conversation state is saved when the ad experience is started, and the conversation state is returned to the state before the advertisement experience was delivered when the ad conversation is ended.

Some embodiments hold conversation state variables on the server. Some embodiments hold conversation state variables on the client device. Some embodiments transfer conversational states with each human representation and machine response sent between the client and server.

In order to provide a seamless experience for potential customers interacting with one virtual agent through multiple devices, some embodiments share conversational states between different apps or devices.

FIG. 3 is a diagram showing an example of a dialogue between a potential customer on the left side and a conversational device on the right side. First, a potential customer asks, "What's the score of the eagles game." The device performs speech recognition and natural language interpretation according to a large array of field grammars. Some of the discipline grammars are local and some are on remote servers. The device recognizes "Eagles" as the name of the sports team and interprets the inquiry as about sports based on that recognition and the use of the word "score" by the inquiry. The device sends a request to the web API of the live sports information provider. Web API identifies that the Eagles are currently actively playing against the Hawks. Through the API, the provider responds to the device that the score is "5-7 (5 to 7)". The device communicates to potential customers a natural language response that reads "eagles 5 hawks 7 (5 points for Eagles and 7 points for Hawks)".

The potential customer then instructs the device to "text what's up to 555-1234". The device interprets this expression and understands that it has a high score in the SMS field. The device parses this expression using a call to the SMS Web API provider, where "what's up" is a text message and "555-1234" is where the text is sent. Judge that it is the telephone number of. The web API provides an acknowledgment to the device, and the device provides a potential customer with a natural language response of "message sent".

The potential customer then asks, "Where can I get some toothpaste." The device interprets this as a high score in the shopping arena, reads the location from the geolocation service software feature, and sends the location information to the provider of the map showing the geographic location of the retail store. The device then accesses the API and looks for retail stores, in this case, Star-Mart, which acknowledges that there is toothpaste in stock, in order of distance from the device's geographic location. The device then provides a natural language response of "there's toothpaste at Star-Mart".

The potential customer then asks, "will it rain tomorrow." The device interprets this as a high score in the field of weather, accesses the weather information provider API to get the weather forecast for the area, and replies "no". The potential customer then said, "what's the
capital of France. " The device interprets this as a question about Wikipedia, sends the Wikipedia API a request for capital information contained in a French Wikipedia article, and gets a response of "Paris". The device provides a natural language response of "Paris is the capital of France".

During this conversation, the device creates an array of conversational variables and values. As a result of the first query, the system adds the variable sports_team with the value Eagles and the variable sports_request with the value score to the conversational state. It is the information that the system uses to access the Sports Web API and get a response. The response consists of two pieces of information that the device stores in a conversational state. Specifically, a variable called an opponent having a value of Hawks and a variable score_va having a value of 5-7.
It is information called blue. As a result of the second exchange, the device adds text_message = "what's up" and text_number = 555-1234 to the conversation state variables. By the third exchange, the device stores the shopping_product = toothpaste, the shopping_request = nearest_store, and the shopping_response = star-mart. By the fourth exchange, the device adds weather_request = will_rain, weather_time = tomorrow, and weather_response = no_rain to the conversational state. By the fifth exchange, the device adds wickedia_request = capital_city, wickedia_article = France, and wickedia_answer = Paris to the conversational state.

The device also holds an array of bid functions for various ad units. The array contains three bidding functions, one bidding function for each of the ad unit for StarShoes, the ad unit for StarPaste, and the ad unit for StarStar. The ad bid function contains the amount offered to serve the ad unit. Some embodiments accept currency-based bids for 1000 ad unit deliveries. Some embodiments tend to support and process more complex ad bidding functions than the ad bidding functions shown in FIG.

The conversational state includes the variable shopping_produc, but since it is not = shoes, the StarShoes ad bid is for 0.3 amount. Since the conversation state includes the shopping_product variable and its value is toothpaste, the StarPaste ad bid is 0.9 amount. The conversation state includes the wikipedia_request variable, but since its value is not person_age, the StarTheater ad bid is 0.8 amount. Therefore, StarPaste is an ad unit with the highest bid function value.

After a certain period of silence, the device delivers an introductory message to the StarPaste ad unit. The introductory message says, "did you know that StarPaste is on sale at Star-Mart for just 2 bitcoins?" It is a natural language voice generated by. This involves potential customers in this ad.

FIG. 4 is a diagram showing a process according to the embodiment. In step 41, the publisher receives the expression and interprets it according to an array of grammar 42 and conversational state variables 43. Therefore, publishers create intents. At stage 44, the publisher takes the action specified by the intent. References to specific entities are also created during the interpretation stage. The entity is stored by the publisher in an array of conversational state variables 43. On the other hand, in stage 45, a plurality of ad units are received. The bid of the ad unit is analyzed against the background of the conversation state 43. Therefore, the analysis in stage 45 selects the highest bidder ad unit and outputs its referral message. The purpose is to involve the lead in making expressions that match the grammar of the ad unit.

When you come across a mood ad, the potential customer has a feeling for the product or service in the ad, depending on the potential customer's emotions when they come across the ad. Emotion detection is the best and most instantaneous measure of a potential customer's current sentiment. There are many emotion detection algorithms that utilize natural language, speech prosody, camera input, and other forms of sensors and algorithms for estimating people's emotions. Affect is the nature of a person, while mood is the nature of the interaction between people or between people and machines.

Some embodiments detect the mood and conditionally deliver an ad unit referral message based on the mood. Some embodiments support mood-conditioned advertising content, such as referral message content. Some embodiments support ad bid values conditioned by the mood. Some embodiments support threshold-based conditioning and some support continuous value-based conditioning. Some embodiments frequently reassess the mood, and some embodiments reassess the mood from time to time.

Conditioning the delivery of ads using the mood gives advertisers control over the sentiment of their potential customers with respect to the products and services they advertise.

Manuscript Conversation Some embodiments of conversational advertising guide potential customers according to the manuscript. In some embodiments, the manuscript may have multiple conditional conversation paths and the machine offers clear options. For example, the machine asks, "would you like to know more about the health benefits, the flavor, or the brightness of StarPaste?" You may.

While the purpose of the referral message is to achieve engagement, the manuscript can encourage potential customers to convert engagement into purchases. Different potential customers have different interests. By analyzing their natural language dialogue, manuscript ad units can classify potential customers. By classifying potential customers, the manuscript can provide information that encourages the conversion of potential customers most effectively.

In some embodiments, the machine uses information steadily gathered from the lead's reaction to profile the momentary state of the lead's thoughts and emotions. For example, if a potential customer speaks faster and louder than usual, the ad may say calming words. As another example, if the conversation state indicates that the recent conversation is about money, the advertisement may provide information about the discounted price.

Knowing how potential customers are involved in advertising can be useful to advertisers and intermediaries. In some embodiments, feedback from the human-machine interface is sent to the publisher, intermediary, or advertiser at each step of the dialogue script. In some embodiments, the feedback indicates the position of the manuscript. In some embodiments, the feedback includes a transcribed or audio recording of the captured potential customer representation. Some embodiments include personally identifiable information. In some embodiments, the advertiser or intermediary is charged a premium for further information.

FIG. 5 is a diagram showing an embodiment of the advertisement dialogue script. The ad dialogue script starts with a referral message 51 that tells the potential customer "did you hear about the new StarPaste?". The interface then waits for the lead's expression to be recognized and matches the received expression with the ad unit syntax. If the expression matches a grammar that indicates indifference, including expressions such as "don't care,""shutup," or silence, the manuscript is "sorry."
The interface device sends feedback on the result to the advertiser.

After the referral message 51, if the lead's expression matches a leading score in areas other than ad 53, the manuscript ends the ad conversation, the interface sends feedback of the results to the advertiser, and the interface ... Proceed to conversations in other areas.

After the referral message 51, if the lead's expression matches the grammar that indicates engagement 54, such as "tell me more" or "what about it," the manuscript continues. And "it gives a healthier mouth and more sparkle (
You'll get a healthier mouth and more sparkle) ”to provide a reply content item. The manuscript requires that potential customers be classified as potential customers who are more interested in health than they look and vice versa. The message is information that appeals to potential customers who care about their health (“healthier mouth”) or information that appeals to potential customers who care about their appearance (“more sparkle”). including.

The next expression for a potential customer in Path 55 is a health grammar by saying something like "healthy how?" Or "does it have fluoride?" If there is a match, the manuscript responds with a message to encourage potential health-conscious customers to move forward. The message is "dentist recommended with double fluoride". When path 56 approaches the appearance grammar of the lead's next expression by saying something like "brighter how?" Or "what about breath?" The manuscript replies with a message to encourage potential customers who care about their appearance to move forward. The message is "number 1 in Hollywood for shiny fresh breath".

After taking Route 55 or Route 56, the manuscript went on to say, "Star Paste is on sale now at Star-Mart for just 2 bitcoins." Use message 57 to encourage potential customers to convert to purchases. At this point, the manuscript ends the conversation and the interface sends the corresponding feedback to the advertiser.

Some embodiments support much more complex manuscripts with many conditional paths, multidimensional lead generation, and conditional personalized message content, and some ad units have the manuscript Prescribe.

Privacy Some embodiments may deliver ads that annoy or embarrass potential customers when others hear or see the introductory message of the ad. For example, some ads can be ads about gifts for a surprise party, some ads can be ads about products that treat an individual's disease, and some ads have an amorous hobby. It can be an advertisement about the service it provides.

In some embodiments, advertising or advertising bidding can be conditioned by the presence of someone other than the target potential customer. Some embodiments perform this conditioning with a camera and image processing for detecting a person. Some embodiments use an audio-speech discrimination algorithm to detect the presence of multiple speakers. Some embodiments detect the transmission of radio signals from personal portable electronic devices such as Bluetooth®, Wi-Fi, or mobile communication network signals from mobile phones.

Some embodiments that detect the presence of another person lower the privacy level variable in the presence of another person. Some such embodiments utilize image processing or voice characterization to detect the attributes of a person who is present, such as determining whether a man or woman, or an adult or child is present. ..

Some embodiments raise the privacy level variable if the human-machine interface in natural language is through a personal interface, such as a headphone, wearable device, or personal mobile phone terminal.

Some embodiments parse the ad unit code to detect the privacy level requirements associated with the ad unit. The embodiment only displays an ad, and in some embodiments, if the current privacy level is above the level specified in the ad unit code, only the ad will be considered when bidding.

Intents and Actions Natural language understanding systems interpret expressions to determine intents. An intent is a data structure that can represent the action that the speaker wants the machine to perform. In some embodiments, the interpretation follows context-free grammatical rules, such as those created by system designers. In some embodiments, the interpretation is made by applying a statistical language model to the expressed words. This can be done in a variety of ways, including by utilizing trained neural networks.

According to some embodiments, the interactive natural language advertisement comprises a grammar having rules that prescribe specific intents according to the words expressed by the potential customer. For example, the grammar may recognize the word "tell me more" as an intent requesting additional information. This means requiring the interface to take action to deliver more information about the product or service of the advertisement. The grammar may recognize the word "shut up" as an intent requesting an action to end an advertising conversation. Even if the grammar for mobile phone advertising recognizes the phrase "how's its battery life" as a request for specific information about the battery capacity and power consumption of the phone in the advertisement. good. The grammar may recognize the word "get it" as an intent requesting an action to order an advertised product or service.

Many means are available to perform the actions specified by the intent of natural language grammar. For example, a text message sending function provides a means of sending a text message. The ordering function provides a means of ordering the delivery of an advertised product or service. Map functions provide a way to tell potential customers how to get to where they can buy an advertised product or service. The browser provides a means for presenting web content such as HTML (Hypertext Markup Language) and CSS (Cascading Style Sheet) for the goods or services advertised. The browser also provides a means for executing a particular type of manuscript, such as a manuscript written in the Javascript® language. Such manuscripts can cause the computer to perform any action that the computer can do. Self-driving vehicles provide a means of transporting potential customers to a place to purchase advertised goods or services. Various means of appropriately performing the actions indicated by the intent for each different application embodiment will be apparent to those skilled in the art.

Various embodiments perform the actions indicated by the intent to the extent possible. Some embodiments respond to potential customers, advertisers, or both if they are unable to perform the intended action.

Bidding According to some embodiments, inventory is the advertising space / time that a publisher can sell. There are many known ways to place advertising bids on available inventory.

FIG. 6 is a diagram showing an embodiment in which advertisers 62a, 62b, and 62c provide an advertising unit including bids and money to publisher 61. The publisher provides the desired content and advertisement to the potential customers 63a, 63b, 63c. Hopefully, ad conversions will occur and potential customers will spend money on goods or services from advertisers.

A bidding system can include a large number of intermediaries and many types of intermediaries that exist between the advertiser and the publisher. FIG. 7 is a diagram showing a system in which advertisers 72a, 72b, and 72c provide ad units to media buyer 75. Media Buyer 75 is a person who collects and packages ad units for distribution to multiple publishers and negotiates pricing. A typical media buyer is a large advertising agency. The media buyer 75 provides the advertising unit to the trading desk 76, which bids on the ad exchange 77. Large advertising agencies tend to have trading desks. Ad Exchange 77 matches ad unit bids with inventory and creates a market with programmatic auction pricing. AppNexus® is an example of an ad exchange. In another route, another advertiser 72d and media buyer 75 provide ad units to ad network 74. Google® AdWords® is an example of an advertising network. The ad network facilitates automatic ordering of ads in available inventory. Publisher 71 provides inventory on Exchange 77 and Advertising Network 74 and delivers it to potential customers 73a, 73b, 73c.

Some ad exchanges carry out programmatic bidding. Program bidding brings high liquidity to the advertising market. Programmatic bidding in various embodiments is included in the user profile, location, location history, current activity, time, day of the week, time of day, type of recently delivered advertisement, type of information contained in the conversation state, and conversation state. It depends on the specific value of information, mood, emotions, and neural activity patterns.

According to some embodiments, the publisher maintains a conversational state about a conversational interface between a human and a machine. Publishers analyze bids for ads in relation to the conversational interface between humans and machines. The bid indicates the interest value of the conversation state variable and is affected by at least one current conversation state variable having the interest conversation state variable value.

Some embodiments support bidding based on the presence of keywords contained in the conversational state. Some embodiments support bidding based on the presence of a particular area of conversation contained in the conversational state. For example, weather is one area and sports is another. Some embodiments support bidding based on the presence of certain variables in the conversational state. For example, sports_team is a variable. Some embodiments support bidding based on the specific value of the variable. For example, "eagles" can be the value of the variable sports_team. Some embodiments support bidding based on a set of possible values for the variable, such as "eagles", "hawks", and "falcons" as the value of the variable sports_teams. .. Some embodiments support bidding based on a range of values for numerical variables, such as temperature variables that have values between 20 ° C and 30 ° C (68 ° F to 86 degrees Fahrenheit). Some embodiments support bidding based on a combination of the above criteria. Some embodiments support specifying a bid as a programmatic expression of any of a variable type, a specific value, a set of values, and the existence of a range. Such an expression determines the degree of agreement between the bid and the conversational state. The formula can also represent an adverse effect on the bid value based on the factors described above.

Some embodiments support bidding based on mood. Such an embodiment detects the current mood value at the human-machine interface and evaluates the bid against the background of the mood value. Some embodiments accept an externally calculated mood value and use this value to influence the calculated bid value.

Some embodiments support bidding conditioned by whether the potential customer is in a private listening environment. Such an embodiment detects this environment and calculates bids accordingly. Some such embodiments determine the listening environment by detecting the presence of others.

In some embodiments, voice activity is detected at the interface between the human and the machine, and the advertisement introduction message is output only when there is no voice activity. Some embodiments wait for a certain amount of time after not detecting voice activity before outputting the ad referral message.

Some embodiments support bidding based on specific intents of natural language expressions. Some embodiments support bidding based on specific actions triggered by one or more intents.

In some embodiments, the advertiser is charged for the bid only when engagement occurs. Some embodiments only charge the advertiser when a conversion occurs. Some embodiments charge the advertiser based on how many times the lead has interacted with the ad prior to conversion or release from engagement. Some embodiments are based on what kind of interaction the lead is interacting with, such as whether the lead makes a price request or whether the lead expresses negative emotions. Mainly charge the price.

Some embodiments are methods of defining bids and methods of analyzing bids. Some embodiments are systems that perform bid analysis. Some embodiments are non-temporary computer-readable media that store bids or bid functions.

Feedback and Analytics Knowing the current state of mind of a potential customer, the very topic they are thinking about, and the specific target of the potential customer's thoughts is of great value to offline analytics by advertisers and analytics companies. However, it is much more valuable to know such information in real time.

It's always difficult, complex, and inaccurate for click-based advertising feedback to determine the psychological state, topics, and thoughts of potential customers. Natural language processing of potential customer communication, such as by processing the content of email messages, is useful in determining psychological states, topics, and objects of thought. However, this is not real time.

Some embodiments, by processing conversations in natural language in real time.
Extract information that is of great value to ad targeting. Such information is useful when applied to real-time bidding algorithms for conversational ad units. Also, such information is useful to advertisers when fed back from human-machine interactions.

Some embodiments feed back information for real-time bidding and create intent data structures by interpreting natural language representations from potential customers. Then, the intent data structure is analyzed and the analysis result is derived in real time. Next, the analysis result is provided to the advertiser's real-time bidding.

Some embodiments provide feedback on an expression-by-expression basis, such as whether engagement or conversion has occurred. Some embodiments provide feedback for each conversation in the advertisement.

In some embodiments, the analysis results include display of intents. Some embodiments with analytical results feed back an indication of the field of conversation prior to engagement. Some embodiments with analytical results feed back the display of the field of conversation immediately after the advertising interaction.

In some embodiments, personally identifiable information is fed back to the advertiser. Some embodiments specifically hide personally identifiable information. One of the reasons for doing so is to guarantee the privacy of system users. Another way to use personally identifiable information is to sell this information to advertisers for a higher payment than you pay for anonymous lead interaction analytics.

Some embodiments include topics, intents, semantics, moods / emotions, questions asked, engagement rates, other engagement metrics, ad conversion rates and ad categories, reach, delivery, and amount spent. Use feedback analytics for, and other conversion metrics. Analytics may include which / how many ads in an array of competing ad bids match the terms of ad unit delivery. This improves the ability to determine when to deliver an ad.

Procter & Gamble (registered trademark) defines FMOT (First Moment Of Truth) as when a potential customer purchases a product, and SMOT (Second Moment Of Truth) as when a potential customer first uses the product. It is defined. Google® defines ZMOT (Zero Moment Of Truth) as the first time a potential customer searches for information about a type of product. Some key query expressions indicate ZMOT, such as "Show me ..." or "Where can I get ...". Some intents derived from natural language interpretation indicate ZMOT. Many language patterns can match the grammar that creates intents related to research on a product or service.

Some embodiments identify advertising opportunities by monitoring natural language interactions in human-machine interfaces. The embodiment interprets natural language representations to identify ZMOT intents. At such times, such an embodiment identifies the type of product or service mentioned in an inquiry or conversation, and upon identifying the product or service mentioned, notifies an advertising bid for this product or service. And raise the ad delivery price because of its timely relevance. Some such embodiments subsequently deliver the ad unit identified by the bid. Some such embodiments deliver an introductory message to the ad unit of a product or service. Some such embodiments subsequently receive expression replies from potential customers and match them to the grammar.

FIG. 8 is a diagram showing a timeline of information flow between an advertiser, an intermediary, a publisher, and a potential customer according to an advertising conversation scenario of the embodiment. At time t0, the advertiser provides the ad unit to the intermediary before the conversation begins. The intermediary determines the price of the ad unit based on the bid on the ad unit in the auction process.

At time t1, which can be well after time t0, the publisher identifies the ad unit with the winning bid. Specifically, publishers determine at least one ad unit based on representations from potential customers. In one embodiment, the publisher determines an ad unit that contains a grammar that matches the expression from the potential customer. Each ad unit contains one grammar and one or more bids. Each bid is associated with at least one intent. The publisher uses the grammar contained in each ad unit to determine at least one intent contained in the expression from the lead. The publisher then selects one bid from one or more bids for the determined ad unit that matches the representation from the potential customer and determines dynamic pricing information based on the selected bid. do. In one example, for an ad unit, the publisher would say, "[can you | please] text [me] a link [to me]. The bid for the intent corresponding to "Show" is decided to be higher than the bid for the intent corresponding to "Show". In another embodiment, each ad unit comprises a discipline-defining grammar. The publisher determines the area of expression defined by the grammar of the determined ad unit. The publisher then decides to bid on the determined ad unit and determines dynamic pricing information based on the determined bid. In one example, for an ad unit, the publisher decides to bid higher on the "weather" field than on the "sports" field. The publisher then provides meta information to the intermediary and advertiser at time t2. Meta information is dynamic, such as winning ad ID, conversation status, most recent conversation area, and mood when choosing an ad, potential customer geographic location and user profile, and whether there is a ZMOT ad serving fee. May include pricing information.

Immediately thereafter, at time t3, the publisher delivers an introductory message to the ad unit. In this scenario, the potential customer provides a voice representation of engagement, also known as spoken click (slick) when used in browser advertising terminology. Slick requests specific information about the pricing of the products in the advertisement. At time t5, the publisher sends a web API request for pricing information to the advertiser. At time t6, the advertiser provides the web API response to the publisher along with the markup text for the TTS response. At time t7, the publisher uses the TTS software built into the publisher's human-machine interface device to synthesize voice into typical machine voice and outputs the synthesized voice from the speaker.

At time t8, the potential customer offers a second slick and asks to place an order. At time t9, the publisher makes a web API request to the intermediary to fulfill the order. The intermediary who fulfills the order processes the order to deliver it and provides the publisher with a web API response at time t10. At time t11, the publisher takes the action of emailing the receipt of the order and provides the potential customer with a TTS response confirming the order.

At time t12, the publisher reports the conversion to the advertiser for use in analytics, and the publisher outputs a thank-you message as TTS voice.

"Boiler plate"
Various embodiments have various arrangements of human-machine interface devices and one or more connected servers for publisher, advertiser, and intermediary roles. In one embodiment, one device communicates with an ad server that includes at least one ad unit.

In another embodiment, the advertiser has multiple divisions in different offices, some advertisers with programmers developing grammar, others with artists developing advertising content. If so, some advertisers have an ad department that combines grammar and content into ad units and combines ad units into campaigns. The advertiser delivers the advertisement to the advertising agency, which stores the advertisement on the database server and accesses the advertisement by a media buyer who interacts with other intermediaries. Media buyers deliver ads and bids to the advertising network and to the third party trading desk that manages the supplier network. The trading desk delivers advertising bids to the Third Bertie Exchange, which matches the bids to inventory. Winning ad units are sent through the device to publishers with geographically dispersed servers to access potential customers with short latency. The ad unit is stored locally on different types of devices designed by different consumer product manufacturers, and the ad unit content is delivered to potential customers under the control of the publisher.

FIG. 9 is a diagram showing an embodiment. The potential customer 91 communicates with the publisher device 92. The publisher device 92 includes a speaker 93 that delivers audio content to potential customers 91. The publisher device 92 further includes a microphone 94. The microphone 94 receives audio including natural language voice from a potential customer 91.

The publisher device 92 communicates with the advertisement server 96 through the network 95. The ad server 96 delivers the ad unit, executes the ad bidding function, and selects the most valuable ad unit to deliver.

FIG. 10A is a diagram showing a non-temporary computer-readable rotating disk medium 101 that stores a computer code. When the computer code is executed by the computer processor, it causes the computer processor to perform the methods and some method steps described herein.

FIG. 10B is a diagram showing a non-temporary computer-readable flash RAM (Random Access Memory) chip medium 102 that stores a computer code. When the computer code is executed by the computer processor, it causes the computer processor to perform the methods and some method steps described herein.

FIG. 10C is packaged with a plurality of computer processor cores comprising components of several embodiments and performing the methods or some method steps described herein by executing computer code. It is a figure which shows the lower (solder ball) side of a SoC (Computer-On-Chip) 103. FIG. 10D is a diagram showing the upper side of the SoC 103.

FIG. 11 shows a rack-based server system 111 used as a component of various embodiments. Such servers are useful as advertiser servers, publisher servers, and servers for the functions of various intermediaries.

FIG. 12 is a block diagram of the core in the system on chip 103. The system-on-chip 103 includes a multi-core computer processor (CPU) 121 and a multi-core graphics accelerator processor (GPU) 122. The CPU 121 and GPU 122 are connected to the DRAM interface 124 and the flash RAM interface 125 through the network on chip 123. The display interface 126 controls the display and is MPEG (Motion Picture).
Allows the system to output Experts Group) video and JPEG (Joint Picture Experts Group) still image advertising content. The I / O interface 127 provides speaker and microphone access for a human-machine interface provided by a device controlled by the SoC 103. Network interface 128 provides access for the device to communicate with the server over the Internet.

FIG. 13 is a diagram showing an embodiment of the server 111. The multiprocessor CPU array 131 and GPU array 132 are connected through a board level wiring 133 to a DRAM subsystem 134 that stores computer code and a network interface 135 that provides Internet access to other servers or publisher devices.

Advertisements in interactive natural language can be placed using various embodiments of the device. Some ads are mediocre, others are intriguing.

FIG. 14A is a diagram showing components of an exciting humanoid robot assistant natural language conversational posting device 11. In order to output audio, the device 11 includes speakers 142 on both sides. The device comprises a microphone array 143. The microphone array 143 includes several MEMS (Microelectromechanical System) microphones that are physically arranged to receive different delay amounts of audio. The device includes an internal processor that executes software that performs DSP (Digital Signal Processing) for detecting the direction of the sound detected by using the microphone array 143. The device 11 further comprises a module 144 having two cameras for providing stereoscopic images and video captures. Furthermore, to detect the location and relative orientation of one or more potential customers, the DSP software performs neural network-based object recognition for models trained on human shapes. The device 11 further includes a display screen 145 for some ad units that outputs visual ad content such as JPEG still images and MPEG video streams. The device 11 further includes wheels 146a and wheels 146b. Wheels 146a and 146b can rotate independently or in unison, respectively. By rotating in unison, the device can move, such as orbiting around a potential customer. By rotating independently, the device can be turned around to face the potential customer and monitor the movement and activity of the potential customer. The device 11 further comprises a power switch 147 that allows the potential customer to silence the device when it becomes annoying.

FIG. 14B is a diagram showing an embodiment of a home virtual assistant and a music playback device 148. FIG. 14C is a diagram showing an embodiment of a Bluetooth compatible earphone device 149. FIG. 14D is a diagram showing an embodiment of a common mobile phone 1410. FIG. 14E is a diagram showing an embodiment of the automobile 1411.

Some embodiments, such as the Home Virtual Assistant 148, give the user little privacy. Some embodiments, such as earphones 149, provide excellent privacy to potential customers. Some embodiments, such as the mobile phone 1410, have a display device screen. Some embodiments are screenless, such as earphones 149, and do not have a display screen. Some embodiments, such as the Home Virtual Assistant 148, are stationary. Some embodiments, such as the car 1411, are mobile. Some embodiments, such as the mobile phone 1410, are stationary.

The method and system of ad delivery and the capabilities of the computer-readable medium make up the complete embodiment, but some elements of the ad flow may be in different countries. That is, passing an ad unit or its components through servers in different countries does not prevent direct infringement of claimed methods, systems, and computer-readable media.

Code Example The following exemplary ad unit code listing is a code listing for an exemplary ad unit. Use a dedicated programming language that uses the same syntax as C language. The bidding function is described on the 3rd to 55th lines. The bid value assigned in the 40th to 53rd lines is conditioned on whether or not it is blocked for a specific user ID. Bid values are also subject to a level of privacy that is personal, as opposed to shared or public. Bid values are further conditioned by mood values that are greater than the threshold. Bid values are also positively affected by the presence of certain recent areas specified in lines 6-8 in the conversational state. The bid value is negatively affected by the presence of a particular field in the conversational state, specified on lines 9-13. Bids are further positively and negatively affected by the presence of the keywords specified on lines 15-27 in recent conversational states. Bids are also positively and negatively impacted by the presence of specific meta-information that describes the current potential customers, environment, and machine state. The extent to which different disciplines, keywords, and meta information affect bids is scaled to 2.0, 1.5, and 2.5, respectively.

Introductory messages are described on lines 61-68. The introductory message includes a display image, an audio clip for output, and a reference to some text for reading that is marked up to emphasize the word.

The ad unit is limited to content that is not offensive at line 71. Ad units are assigned and line 79 provides a high level of information that publishers report to advertisers. The ad is configured to wait 10.0 seconds for the lead's reaction before considering ending the conversation without engagement on line 82.

Lines 85-96 define a grammar that includes a particular phrase with optional words. A grammar intent is a function call to a function specified outside the ad unit or a function specified inside the ad unit. In each line, the intent is associated with the grammar. As an example, on line 87, "can you | please] text [me] a link [to
"me]" is the part that defines the grammar, and "text_url ()" is the part that defines the intent of this expression and the action of the device 11. As another example, in line 94, "i don't care" is the part that defines the grammar, and "set_dont_care ()" is the part that defines the intent of the expression and the device. This is the part that defines the 11 actions. If the device 11 determines that the natural language expression received from the potential customer matches one or more of the grammars on lines 85-96, the device 11 determines that the intent associated with the matching grammar is that natural. Judge that it is an intent of linguistic expression.

Lines 99-119 specify content elements that include video clips, animations, images, video streams, and TTS text marked up with conditional content such as finding the city closest to the latitude and longitude of the current device. NS.

Lines 122-167 define custom functions called by the grammar intent. Lines 123-132 specify a function for delivering a sequence of additional information content in response to a potential customer's request for more information. Lines 134 to 137 specify a function for sending a text message. Lines 139-142 specify a function for sending an e-mail message. Lines 144-147 specify a function for checking the price of a particular SKU (Stock Keeping Unit) number of the product in the advertisement. In lines 149 to 152, when the posting device has a display screen, a function for displaying an image of the product is specified. Lines 154 to 164 specify the nearest store where the SKU is available, the price of the product at that nearest store, and a function to find out if the product is currently on sale.

On lines 165 to 167, a function is specified for setting a display for each user ID as to whether or not the potential customer has indicated that he / she is not interested in the product. If a potential customer indicates that they are not interested in the product, the ad will no longer be served to that user ID.

The code shown in the example below is only an example of some features of some embodiments. Advanced advertisers create ad units with more code, more sophisticated bids, more complex and diverse grammars, more dependencies, and more and more complex content. Some embodiments provide other system functions available for use in ad units to support more complex code configurations.

Some embodiments provide simple templates for less sophisticated advertisers. For example, embodiments provide SKU and TTS strings to provide pricing and location capabilities, web content display capabilities, and a grammar for answering general product inquiries. Automatically provide the function of. The advertising rules for simple templates are as follows.

BID = 1.5
SKU = 9781542841443
INTRO = "did you hear about the new Star Paste?"
MORE = "it gives a healthier mouth and more sparkle"
The system provides a grammar that includes responding to "tell me more" with a pricing search using the TTS and SKU values of the MORE text.

Illustrative ad unit code listing
1 ad StarPaste {
2 float PPKI = 0.25; // Default price per 1000 impressions
3 bid {
4 float mood_min: 0.5; // Lower mood threshold
5 // Conversation weighting field
6 domain cntxt_domain_pos = {"health",
7 "grocery",
8 "dating"};
9 domain cntxt_domain_neg = {"math",
10 "geography",
11 "MoonPaste", // Competing product ads
12 "Moon Paste",
13 "Moon Paste"};
14 // Keyword weighting
15 words cntxt_words_pos = {"smell",
16 "breath",
17 "breath", // increase weight by repeating
18 "breath",
19 "fresh",
20 "date",
21 "night",
22 "sleep",
23 "morning"};
24 words cntxt_words_neg = {"directions",
25 "text",
26 "call",
27 "problem"};
28 // Meta information weighting
29 meta cntxt_meta_pos = {"location = US",
30 "gender = female",
31 "age>16",
32 "environment = home",
33 "environment = retail-grocery",
34 "people_present <4",
35 "male_present = FALSE"};
36 meta cntxt_meta_neg = {"time = 9to5",
37 "environment = restaurant",
38 "male_present = TRUE"};
39 // Don't serve ads to potential customers who are in a bad mood
40 float bid_val =
41 check_dont_show (user_id ())? 0:
42 (privacy_level <PERSONAL)? 0:
43 (mood <mood_min)? 0:
44 PPKI * mood
45 // Only if the recent conversation area contains more positive areas than negative areas
46 // Deliver ads
47 * 2.0 * min (count (cntxt_domain_pos) --count (cntxt_domain_neg), 0))
48 // Only if recent keywords contain more positive areas than negative areas
49 // Deliver ads
50 * 1.5 * min (count (cntxt_words_pos) --count (cntxt_words_neg), 0)
51 // Only if recent meta information contains more positive areas than negative areas
52 // Deliver ads
53 * 2.5 * min (count (cntxt_meta_pos) --count (cntxt_meta_neg), 0);
54 return bid_val;
55}
56
57 // The introductory message is
58 // If the human-machine interface has a screen, use an image,
59 // Use tts if the human-machine interface has a TTS service, 60 // Use audio if the tts service is not available.
61 intro {
62 image = "./media/StarPaste.jpg"; // Reference to image content
63 // For devices with audio output but no TTS function
64 audio = "./media/jingle.mp3";
65 // The system outputs audio from the global tts string variable
66 tts = "have you heard about the <emphasis level =" strong "> new
67 </ emmphasis> Star brand Star Paste? ";
68}
69
70 // Global settings to block ads with offensive content or grammar
71 non_offensive_restriction = TRUE;
72
73 // On the system,
74 // Each conversational expression (conversational expression that hits the grammar, 75 // if it does not hit another grammar, the conversational expression immediately after), user ID, time stamp, position,
76 // Language / Accent, Existence of person (if detected, specific ID, or
77 // Specific numbers and characteristics), mood, conversational status,
78 // Global settings to let advertisers report
79 reporting_level = HIGH;
80
81 // Global setting for the number of seconds before timeout when there is no response
82 conversation timeout = 10.0;
83
84 // Grammar
85 grammar {
86 "tell me more"=> next_more ();
87 "[can you | please] text [me] a link [to me]"=> text_url ();
88 "[can you | please] send [me] a link [to me]"=> email_url ();
89 "how much is it"=> say_cost ();
90 "[how much | what] does it cost"=> say_cost ();
91 "show me"=> show_product ();
92 "where can I [buy | get] [it | some]"=> where_to_buy ();
93 "i don't care"=> set_dont_care ();
94 "dont tell me about it"=> set_dont_care ();
95 "i hate [it | that]"=> set_dont_care ();
96}
97
98 // Content objects defined for screens and screenless devices
99 content more1 {
100 tts = "it's made with ground unicorn horn for a magic sparkle";
101 video = "./media/sparkle.mpg"; // Reference to video content
102}
103 content more2 {
104 tts = "use it today and get a date by tonight, guaranteed";
105 flash = "./media/date.swf"; // Reference to Flash animation
106}
107 content more3 {
108 tts = "it's <nearest_city (lat_long ())>'s silkiest tooth paste";
109 image = "./media/lingerie.gif"; // Reference to silk GIF animation
110}
111 content more4 {
112 tts = rand (
113 "why don't you add it to your cart",
114 "ask your shopping assistant for express delivery",
115 "it's available everywhere that fine toiletries are sold",
116);
117 // Reference to live stream of celebrity spokesperson
118 stream = "http://www.starpastes.com/celebrity.strm";
119}
120
121 // Grammar can read ad-specific functions
122 int more_idx = 0;
123 function next_more {
124 if (more_idx <= 4) more ++;
125 case (more_idx) {
126 // The deliver function handles the output of the content object
127 1: deliver (more1);
128 2: deliver (more2);
129 3: deliver (more3);
130 4: deliver (more4);
131}
132}
133 // Send SMS
134 function text_url {
135 send_text ("http://www.starpastes.com/robo_ad.html");
136 tts = "here you go";
137}
138 // Send an email
139 function email_url {
140 send_email ("http://www.starpastes.com/robo_ad.html");
141 tts = "done";
142}
143 // Say the price
144 function say_cost {
145 float price = get_price ("9781542841443"); // StarPaste SKU
146 tts = "it's only <price>";
147}
148 // Display the product on the screen
149 function show_product {
150 if (DISPLAY_SCREEN = TRUE)
151 image = "./media/StarPaste.jpg";
152}
153 // Say where to buy the item, the price, and if it's a sale
154 function where_to_buy {
155 // Find the nearest store with items in stock
156 int store_id = nearest_stock ("9781542841443");
157 // Check the price at the nearest store using the store_name function
158 tts = "it's available at <store_name (store_id)> for just
159 <price_check (store_id, "9781542841443")>";
160 // Use the store_sales function to see if an item is on sale
161 if (store_sales (store_id, "9781542841443"))
162 tts = "it's on sale at <store_name (store_id)>
163 <emphasis level = "strong"> today </ emphasis>";
164}
165 set_dont_care {
166 set_dont_show (user_id ());
167}
168}

Claims (19)

It is a data structure of an ad unit, and the ad unit is
A display showing the content of the introductory message and
Includes natural language grammar provisions associated with each of the ad units, the natural language grammar contains at least one intent.
If the ad server system interprets an expression from a potential customer that matches the natural language grammar after the referral message has begun to be delivered, the natural language grammar is used by the ad server system and the natural language is used. A data structure that determines the intent included in the grammar of the above as the intent of the expression. A step of specifying a bid function for an ad unit, said bid function, when executed by a system storing the value of a conversational state variable representing a human-machine conversation, said the stored value of the conversational state variable. A method that allows different bid values to be calculated. The method of claim 2, wherein the conversation state variable represents one or more keywords. The method of claim 2 or 3, wherein the conversation state variable represents a field. The method according to any one of claims 2 to 4, wherein the bid value further depends on the result of the equation by the program. The method according to any one of claims 2 to 5, wherein the calculation of the bid value differs depending on the value of the mood variable. The method of any one of claims 2-6, wherein the bidding function is conditioned by a display that desires privacy. A program that, when executed by a computer, causes the computer to receive an expression from a potential customer and make a real-time bid decision based on the intent of the expression. A program that, when executed by a computer, causes the computer to receive an expression from a potential customer and make a real-time bid decision based on the area of the expression. A way to feed back information about real-time bidding
Steps to interpret natural language representations from potential customers to create intent data structures,
The step of analyzing the intent data structure and deriving the analysis result in real time,
A method comprising providing the analysis results to an advertiser's real-time bidding. The method of claim 10, wherein the analysis results include an intent display. The method of claim 10 or 11, wherein the analysis results include an indication of the field of conversation before the potential customer makes the natural language expression. The method according to any one of claims 10 to 12, wherein the analysis result includes display of the field of conversation immediately after the advertisement interaction. The method according to any one of claims 10 to 13, wherein the analysis result includes personally identifiable information. The method of any one of claims 10-14, wherein the analysis hides personally identifiable information. It ’s a way to identify advertising opportunities.
Steps to monitor natural language dialogue in human-machine interfaces,
A step of interpreting a natural language expression as a query that identifies a ZMOT (Zero Moment Of Truth) intent,
Steps to identify the type of product or service mentioned in the inquiry,
Methods, including steps to notify ad bids. 16. The method of claim 16, further comprising the step of delivering the ad unit identified by the ad bid. 16. The method of claim 16 or 17, further comprising delivering a referral message associated with the ad bid. 18. The method of claim 18, further comprising the step of receiving a grammatically consistent reply.

Images