Patent Application
20180068012
This disclosure relates generally to dialog computer systems, and more specifically, to chat flow tree structure adjustment based on sentiment and flow history.
A dialog system can provide a chat flow to a user to accomplish a task. A dialog system can often have hundreds of chat flows within it which range from welcoming the user, gathering basic user info, performing tasks for the user, and the like. These dialog systems are also being used in combination with natural language question answering systems, where a chat flow can provide a natural language response to a question asked within the chat flow.
Gathering user feedback within a dialog system is inherently difficult. Typical feedback mechanisms of a 5-star rating system or thumbs up/thumbs down feedback do not integrate well in a natural language conversation interface. Thus, it can be challenging for dialog system operators to know whether chat flows need modification and to accurately identify specific portions of the chat flows to improve.
According to embodiments, a method, system, and computer program product are provided. A chat flow tree structure is accessed that defines a plurality of response sequences for a chat flow of a dialog system. A sentiment and flow history associated with the chat flow tree structure is analyzed. The sentiment and flow history defines an observed user sentiment captured for one or more system responses of the response sequences. Each of the one or more system responses is flagged having a negative sentiment in the sentiment and flow history and a sample size greater than a minimum adjustment threshold. A portion of the chat flow tree structure is adjusted that includes at least one of the flagged system responses to reduce the negative sentiment.
Additional features and advantages are realized through the techniques of the invention. Other embodiments and aspects of the invention are described in detail herein and are considered a part of the claimed invention. For a better understanding of the invention with the advantages and the features, refer to the description and to the drawings.
The subject matter which is regarded as the invention is particularly pointed out and distinctly claimed in the claims at the conclusion of the specification. The forgoing and other features, and advantages of the invention are apparent from the following detailed description taken in conjunction with the accompanying drawings, which:
Embodiments use facial expression tracking, voice characteristics, and/or interaction pattern observation to monitor real-time user sentiment toward responses returned by a dialog system. A path taken by the dialog system can then be altered to respond to emotional responses of a user without any direct input from the user to evaluate the system responses. Real-time user feedback as user sentiment can be used (e.g., by way of facial expression tracking, voice characteristics, and/or interaction patterns) to reroute a conversation to different conversation nodes within the dialog system. Complex dialog flows are inherently difficult to script. In embodiments, the dialog system can have an additional layer of feedback used to help calculate when the user has suddenly changed intent or when the system has responded in a way which surprised or frustrated the user. A planned sequence of the chat flow can be rerouted from a next system response to a sentiment-based repair sequence to alter content delivered to the user based on a detected aspect of the user sentiment. Embodiments can also track sentiment and flow history to identify and adjust system responses from a portion of a chat flow tree structure to reduce negative user sentiment.
Turning now to
In the example depicted in
In embodiments, the sentiment monitor 112 executes on the dialog system 102 by receiving an image stream or video feed from the user systems 104A-N along with text and/or audio. In alternate embodiments, a portion or all of the sentiment monitor 112 runs on the user systems 104A-N, for instance, as a component of chat flow interfaces 116A-N, and the user sentiment is sent along with other user data (e.g., text/audio user statements, link selections, etc.) to the dialog system 102. In embodiments, a camera 275 (
As the sentiment monitor 112 determines user sentiment, a sentiment and flow history can be captured in sentiment and flow history 128, which may be further partitioned into a short-term history 128A and a long-term history 128B. The sentiment and flow history 128 can define an observed user sentiment captured for one or more system responses of response sequences in a chat flow tree structure 122 for a chat flow 120 of the dialog system 102. A chat flow refiner 114 can use chat flow adjustment rules 118 to analyze the sentiment and flow history 128 and flag each of the one or more system responses having a negative sentiment in the sentiment and flow history 128 and a sample size greater than a minimum adjustment threshold. The chat flow refiner 114 can adjust a portion of the chat flow tree structures 122 including at least one of the flagged system responses to reduce the negative sentiment according to process 800 of
Although only one dialog system 102 is depicted in
In exemplary embodiments, the dialog system 102 and the chat flow refinement system 108 are each implemented on a high-speed processing device (e.g., a mainframe computer system, a cloud computing system, or the like) including at least one processing circuit (e.g., a computer processor/CPU) capable of reading and executing instructions, and handling interactions with various components of the system 100. Alternatively, the dialog system 102 and/or the chat flow refinement system 108 can be embodied entirely in hardware including circuits and memory to perform processes as described herein.
In exemplary embodiments, the user systems 104A-N can include a variety of computing devices with processing circuits and I/O interfaces, such as keys/buttons, a touchscreen, audio input, a display device and audio output. The user systems 104A-N may be embodied in any type of computing device known in the art, such as a laptop, tablet computer, mobile device, personal computer, workstation, server, and the like. Accordingly, the user systems 104A-N can include various computer/communication hardware and software technology known in the art, such as one or more processing units or circuits, volatile and non-volatile memory including removable media, power supplies, network interfaces, support circuitry, operating systems, and the like.
Turning now to
In some embodiments, as shown in
The I/O devices 240 may further include devices that communicate both inputs and outputs, for instance disk and tape storage, a network interface card (NIC) or modulator/demodulator (for accessing other files, devices, systems, or a network), a radio frequency (RF) or other transceiver, a telephonic interface, a bridge, a router, and the like.
The processor 205 is a hardware device for executing hardware instructions or software, particularly those stored in memory 210. The processor 205 may be a custom made or commercially available processor, a central processing unit (CPU), an auxiliary processor among several processors associated with the computing device 200, a semiconductor based microprocessor (in the form of a microchip or chip set), a macroprocessor, or other device for executing instructions.
The memory 210 may include one or combinations of volatile memory elements (e.g., random access memory, RAM, such as DRAM, SRAM, SDRAM, etc.) and nonvolatile memory elements (e.g., ROM, erasable programmable read only memory (EPROM), electronically erasable programmable read only memory (EEPROM), programmable read only memory (PROM), tape, compact disc read only memory (CD-ROM), disk, diskette, cartridge, cassette or the like, etc.). Moreover, the memory 210 may incorporate electronic, magnetic, optical, or other types of storage media. Note that the memory 210 may have a distributed architecture, where various components are situated remote from one another but may be accessed by the processor 205.
The instructions in memory 210 may include one or more separate programs, each of which comprises an ordered listing of executable instructions for implementing logical functions. In the example of
Additional data, including, for example, instructions for the processor 205 or other retrievable information, may be stored in storage 220, which may be a storage device such as a hard disk drive or solid state drive. The stored instructions in memory 210 or in storage 220 may include those enabling the processor to execute one or more aspects of the systems and methods of this disclosure.
The computing device 200 may further include a display controller 225 coupled to a display 230. In some embodiments, the computing device 200 may further include a network interface 260 for coupling to a network 265, such as network 106 of
Systems and methods according to this disclosure may be embodied, in whole or in part, in computer program products or in computing device 200, such as that illustrated in
When a user is traversing a branch of a chat flow tree structure 122 in the dialog system 102 of
In this case, the user is resisting the track (path/branch) proposed by the dialog system 102. The dialog system 102 is trained/programmed to try to obtain all of the address information from the user before moving to the next chat sequence (e.g., obtain a quote). This can present a difficult situation for a typical dialog system to handle. Further, going back a single step in chat flow can present difficulties. For example, if the user asks what her insurance deductible was, the dialog system 102 may respond by asking if the user meant home insurance or car insurance. Accidentally clicking on the wrong link in a response can get the user stuck in a chat flow for the wrong insurance type, as illustrated in the following example.
The user in this example had no easy way to get back to select Home Insurance instead of Auto Insurance, and the dialog system 102 may have no way for the user to directly indicate that the user was unhappy with auto deductible options. Embodiments observe user sentiment based on facial expressions, voice characteristics, and/or interaction patterns to reroute a next system response from a planned sequence of the chat flow to a sentiment-based repair sequence to alter content delivered to the user based on a detected aspect of the user sentiment. Through the use of the sentiment monitor 112 of
The user is unhappy with this response as detected by the sentiment monitor 112. The conversation path selector 110 can use the user sentiment to perform one or more of the sentiment-based repair sequences 124 of
As one example, the sentiment-based repair sequences 124 include augmenting an existing system response with an option to return to the last point in the chat flow with a positive reaction. This option is illustrated in the following sequence.
A further option for the sentiment-based repair sequences 124 includes prompting the user to accept further assistance associated with the system response as illustrated in the following example sequence.
In this example, the dialog system 102 does not wait for the user to explicitly state why she was unhappy; rather, the dialog system 102 preempts the user by applying the sentiment-based repair sequences 124 to provide an improved user experience.
As another example, the sentiment-based repair sequences 124 include waiting for user input as illustrated in the following example sequence.
In this scenario, when the user indicates a negative sentiment and provides a new user statement, a different matching algorithm is used to redirect the sequence within the chat flow. Instead of assuming that the input is related to the current subtree (auto insurance), the dialog system 102 can search the parent tree and sibling subtrees. In this case, a match of “Home Insurance” can be identified from the user statement to shift the chat flow to a “Home Insurance” flow for deductibles and redirects the user to that chat flow instead.
In some instances, users may ask questions that the dialog system 102 does not understand, and this can be frustrating to the user. In this situation, the dialog system 102 can monitor the number of negative reactions by the user as an indication of when to intervene (e.g., a change in a sentiment trajectory beyond a minimum repair threshold). For example, the user may be asking a question about umbrella policies to a system that only knows about home and auto insurance as illustrated in the following sequence.
The dialog system 102 was able to match home insurance with the general “Home Insurance” chat flow in this example. So this appears to be a reasonable response for the dialog system 102. However, the user reaction is negative since she was asking about something completely different.
Again liability coverage would match an Auto Liability chat flow to help a user choose how much liability coverage she would need. In this case, again, the user reaction would be negative.
Having received two negative responses, this could mean different things at different locations within the chat flow, but regardless, the overall negative impression of the dialog system 102 can trigger an action as illustrated in the following sequence.
Here, the user has been redirected to a “negative sentiment” chat flow of the sentiment-based repair sequences 124, which tries to show the user how to use the dialog system 102. A negative sentiment chat flow can give the user information about how the dialog system 102 works, what categories of questions can be asked, or ultimately give the user an option to speak to an agent as the dialog system 102 may not be trained to deal with the question/request. Selectively providing the user with various repair options can help users have a better experience with the dialog system 102.
Over time an overall sentiment per chat flow/sub-flow can be built up in the sentiment and flow history 128. As the complexity and size of the chat flow tree structures 122 grow, it can become difficult to maintain. Questions, responses, and flows can become obsolete. Given difficulties with classic feedback mechanisms, the user sentiment captured in the sentiment and flow history 128 can be examined to determine which parts of the chat flow tree structures 122 are not being well received by users. Chat flows 120, where users generally give up and are unhappy, can be pruned or refactored. By tracking sentiment for every system response, a very fine grained analysis can be performed on the chat flows 120 to identify problem areas.
In another example, some chat flows 120 expect negative responses as illustrated in the following example.
In this case, the answer is correct, but it is likely that the user will not be happy about it. A given flow can be denoted to expect negative responses or surprised responses without triggering redirects. The dialog system 102 can still prompt the user to be sure the answer is correct as illustrated in the following example.
In some embodiments, sentiment data, such as negative sentiment, is tracked with respect to content. Sentiment associations to content can allow content owners to better understand what aspects users consistently show a positive or negative sentiment towards. Such sentiment to content feedback can support product improvement. For instance, consider the following example chat flow with dialog system 102.
Chat flows such as this example can be tagged as a product enhancement request. The level of negativity as well as the number of negative responses may provide a priority to such product enhancement requests.
As the chat flow 502 traverses through the chat flow tree structure 500 based on interactions with a user of one of the chat flow interfaces 116A-N of
Turning now to
At block 604, the sentiment monitor 112 determines a user sentiment associated with a system response provided by the dialog system 102 as part of the chat flow 502 based on observation of the user. For instance, a facial expression, voice characteristics, and/or interaction patterns of the user can be tracked in real-time as the user interacts with the dialog system 102. A camera 275 of the user systems 104A-N can capture the facial expressions as a stream of images or video, and/or microphone 270 of the user systems 104A-N can capture audio data to determine user sentiment from voice characteristics. As one example, the sentiment monitor 112 can apply the process 700 of
At block 606, the conversation path selector 110 can reroute a next system response from a planned sequence of the chat flow 502 to a sentiment-based repair sequence 514 to alter content delivered to the user based on a detected aspect of the user sentiment. The aspect of the user sentiment can include one or more of a surprise sentiment, an anger sentiment, and/or a disgust sentiment. Alternatively, the aspect of the user sentiment can include a change in a sentiment trajectory beyond a minimum repair threshold. For example, if a sentiment trajectory is considered negative and persists for multiple system responses and/or gets progressively more negative over multiple responses, rerouting may be performed where the minimum repair threshold can be defined in terms of a simple count of negative events or an increasingly negative trend of user sentiment over a period of time. The sentiment-based repair sequence 514 can include one or more of augmenting the system response to return to a last point in the chat flow 502 where a positive sentiment was detected, prompting the user to accept further assistance associated with the system response, and/or searching the chat flow 502 to locate another potential match for a user statement that resulted in the system response.
In embodiments, the sentiment monitor 112 can continue monitoring the user sentiment associated with the sentiment-based repair sequence 514 based on the facial expression, voice characteristics, and/or interaction patterns of the user and progressively provide one or more alternate repair options in the sentiment-based repair sequence 514 based on determining that the user sentiment is a negative sentiment during the sentiment-based repair sequence 514. Further, the conversation path selector 110 can determine whether the system response is marked as expecting a negative sentiment and block the sentiment-based repair sequence 514 based on determining that the system response is marked as expecting the negative sentiment.
Turning now to
At block 706, changes in affective states, called sentiment trajectories, of the user are identified based on the facial muscle actions. For instance, facial expressions can be translated into six basic emotions: Happiness, Sadness, Surprise, Anger, Disgust, and Fear. At block 708, eye gazes and three-dimensional poses of the faces are estimated, and gaze directions of the faces are estimated based on the estimated eye gazes and the three-dimensional poses of the faces. At block 710, gaze targets of the audience (e.g., one or more users) are estimated based on the estimated gaze directions and the position of the faces. At block 712, the response of the audience to the dialog content is determined, by analyzing the sentiment trajectories in relation to time and screen positions of a specific dialog.
Turning now to
At block 802, a chat flow tree structure 500 is accessed defining a plurality of response sequences 508 for a chat flow 502 of a dialog system 102.
At block 804, a sentiment and flow history 128 associated with the chat flow tree structure 500 is analyzed, where the sentiment and flow history 128 defines an observed user sentiment captured for one or more system responses 512 of the response sequences 508. At block 806, each of the one or more system responses 512 is flagged having a negative sentiment in the sentiment and flow history 128 and a sample size greater than a minimum adjustment threshold. The minimum adjustment threshold can be used to ensure that adjustments to the chart flow tree structure 500 are not automatically performed in response to a single user experiencing a single problem. The negative sentiment can be one or more of a surprise sentiment, an anger sentiment, and/or a disgust sentiment. Analysis of the sentiment and flow history 128 can include searching for a history of rerouting a next system response from a planned sequence of the chat flow 502 to a sentiment-based repair sequence 514 to alter content delivered to the user based on a detected aspect of the user sentiment. Analysis of the sentiment and flow history 128 can also include searching for the negative sentiment in combination with a premature termination of the chat flow 502, e.g., where a user quits in frustration.
At block 808, a portion of the chat flow tree structure 500 including at least one of the flagged system responses is adjusted to reduce the negative sentiment. Adjusting the portion of the chat flow tree structure 500 can include inserting one or more prompts for the user to accept further assistance prior to one or more of the flagged system responses. Adjusting the portion of the chat flow tree structure 500 can include removing one or more of the flagged system responses from the chat flow tree structure 500.
In some embodiments, the sentiment and flow history 128 includes a short-term history 128A and a long-term history 128B that stores data collected over a longer period of time than the short-term history 128A. The chat flow refiner 114 can compare the user sentiment in the short-term history 128A with the user sentiment in the long-term history 128B after adjusting the portion of the chat flow tree structure 500 to determine whether the adjustment was effective in reducing the negative sentiment. If the adjustment is deemed ineffective, the chat flow refiner 114 may perform another or alternative adjustment (e.g., performing an additional adjustment, or undo the adjustment and try a different adjustment). A product enhancement request can be identified by the chat flow refiner 114 or other analysis tool based on a level of negativity and a number of negative responses associated with content provided to one or more users.
Technical effects and benefits include conversation path rerouting and chat flow tree structure adjustment based on observed user sentiment.
The present invention may be a system, a method, and/or a computer program product. The computer program product may include a computer readable storage medium (or media) having computer readable program instructions thereon for causing a processor to carry out aspects of the present invention.
The computer readable storage medium can be a tangible device that can retain and store instructions for use by an instruction execution device. The computer readable storage medium may be, for example, but is not limited to, an electronic storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination of the foregoing. A non-exhaustive list of more specific examples of the computer readable storage medium includes the following: a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), a static random access memory (SRAM), a portable compact disc read-only memory (CD-ROM), a digital versatile disk (DVD), a memory stick, a floppy disk, a mechanically encoded device such as punch-cards or raised structures in a groove having instructions recorded thereon, and any suitable combination of the foregoing. A computer readable storage medium, as used herein, is not to be construed as being transitory signals per se, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through a waveguide or other transmission media (e.g., light pulses passing through a fiber-optic cable), or electrical signals transmitted through a wire.
Computer readable program instructions described herein can be downloaded to respective computing/processing devices from a computer readable storage medium or to an external computer or external storage device via a network, for example, the Internet, a local area network, a wide area network and/or a wireless network. The network may comprise copper transmission cables, optical transmission fibers, wireless transmission, routers, firewalls, switches, gateway computers and/or edge servers. A network adapter card or network interface in each computing/processing device receives computer readable program instructions from the network and forwards the computer readable program instructions for storage in a computer readable storage medium within the respective computing/processing device.
Computer readable program instructions for carrying out operations of the present invention may be assembler instructions, instruction-set-architecture (ISA) instructions, machine instructions, machine dependent instructions, microcode, firmware instructions, state-setting data, or either source code or object code written in any combination of one or more programming languages, including an object oriented programming language such as Smalltalk, C++ or the like, and conventional procedural programming languages, such as the “C” programming language or similar programming languages. The computer readable program instructions may execute entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer or entirely on the remote computer or server. In the latter scenario, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or the connection may be made to an external computer (for example, through the Internet using an Internet Service Provider). In some embodiments, electronic circuitry including, for example, programmable logic circuitry, field-programmable gate arrays (FPGA), or programmable logic arrays (PLA) may execute the computer readable program instructions by utilizing state information of the computer readable program instructions to personalize the electronic circuitry, in order to perform aspects of the present invention.
Aspects of the present invention are described herein with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and/or block diagrams, and combinations of blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer readable program instructions.
These computer readable program instructions may be provided to a processor of a general purpose computer, special purpose computer, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions/acts specified in the flowchart and/or block diagram block or blocks. These computer readable program instructions may also be stored in a computer readable storage medium that can direct a computer, a programmable data processing apparatus, and/or other devices to function in a particular manner, such that the computer readable storage medium having instructions stored therein comprises an article of manufacture including instructions which implement aspects of the function/act specified in the flowchart and/or block diagram block or blocks.
The computer readable program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other device to cause a series of operational steps to be performed on the computer, other programmable apparatus or other device to produce a computer implemented process, such that the instructions which execute on the computer, other programmable apparatus, or other device implement the functions/acts specified in the flowchart and/or block diagram block or blocks.
The flowchart and block diagrams in the Figures illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. In this regard, each block in the flowchart or block diagrams may represent a module, segment, or portion of instructions, which comprises one or more executable instructions for implementing the specified logical function(s). In some alternative implementations, the functions noted in the block may occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved. It will also be noted that each block of the block diagrams and/or flowchart illustration, and combinations of blocks in the block diagrams and/or flowchart illustration, can be implemented by special purpose hardware-based systems that perform the specified functions or acts or carry out combinations of special purpose hardware and computer instructions.
The descriptions of the various embodiments of the present invention have been presented for purposes of illustration, but are not intended to be exhaustive or limited to the embodiments disclosed. Many modifications and variations will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the described embodiments. The terminology used herein was chosen to best explain the principles of the embodiments, the practical application or technical improvement over technologies found in the marketplace, or to enable others of ordinary skill in the art to understand the embodiments disclosed herein.
