WO2022082453A1

WO2022082453A1 - Artificial intelligence system for transportation service related safety issues detection based on machine learning

Info

Publication number: WO2022082453A1
Application number: PCT/CN2020/122271
Authority: WO
Inventors: Baochang MA; Yulong ZHOU; Kun Han
Original assignee: Beijing Didi Infinity Technology And Development Co., Ltd.
Priority date: 2020-10-20
Filing date: 2020-10-20
Publication date: 2022-04-28
Also published as: CN115698982A

Abstract

An artificial intelligence system for detecting a transportation service related safety issue. The system includes a storage device configured to store a first sample description of an incident and a known safety issue of the incident. The first sample description includes first set of labeled texts. The system further includes at least one processor. The at least one processor is configured to segment the first sample description into word segments, determine word vectors of the respective word segments and determine features of the first set of labeled texts in the first sample description. The at least one processor is also configured to train the learning model based on the word vectors and the features. The system also includes a communication interface configured to provide the learning model for automatically detecting the transportation service related safety issue from an incident description.

Description

ARTIFICIAL INTELLIGENCE SYSTEM FOR TRANSPORTATION SERVICE RELATED SAFETY ISSUES DETECTION BASED ON MACHINE LEARNING

TECHNICAL FIELD

The present disclosure relates to artificial intelligence (AI) systems and methods for detecting transportation service related safety issues, and more particularly to, AI systems and methods for automatically detecting transportation service related safety issues using machine learning.

BACKGROUND

Safety issue detection is usually performed by a transportation service platform to screen events that have safety concerns during transportation services. One way to detect safety issue is based on incidents reported by a passenger or the provider of the transportation service. For example, if the passenger reports that the service provider is acting weirdly such as driving under influence, the service ride may be manually labeled as having a safety issue and the platform will be notified for taking further actions such as instructing the provider to stop immediately, disfranchising the provider’s license to provide service or giving notice to a local police department about the provider’s violation of the law.

Safety issue detection is typically performed by customer service receptionists retained by the transportation service platform manually. For example, a receptionist may pick up a phone call and determine if there is a safety issue based on the incident described by the caller using the receptionist’s personal judgement. In addition, some transportation service platform provides safety concern report options for passengers to notify the customer service when they are under safety threats. For example, DiDi ^TM transportation service application has an “one-key report” feature, where passenger can press one button to trigger a safety report to customer service.

However, the existing methods are not accurate or efficient. For example, using operators to manually detect safety issues based on incident descriptions is expensive and becomes impractical if the amount of calls is huge. In addition, application-based passenger report features cause high mis-alarm rates. For example, passengers tend to press the button when they need customer service assistances, most of which are unrelated to safety concerns. As a result, an automatic and accurate safety issue detection method based on incident descriptions is needed.

Embodiments of the disclosure address the above problems by providing improved artificial intelligence systems and methods for automatically detecting a transportation service related safety issue from incident descriptions using machine learning.

SUMMARY

In one aspect, embodiments of the disclosure provide an artificial intelligence system for training a learning model for detecting a transportation service related safety issue. The system includes a storage device configured to store a first sample description of an incident and a known safety issue of the incident. The first sample description includes first set of labeled texts. The system further includes at least one processor. The at least one processor is configured to segment the first sample description into word segments and determine word vectors of the respective word segments. The at least one processor is further configured to determine features of the first set of labeled texts in the first sample description. The at least one processor is also configured to train the learning model based on the word vectors and the features. The system also includes a communication interface configured to provide the learning model for automatically detecting the transportation service related safety issue from an incident description.

In another aspect, embodiments of the disclosure also provide an artificial intelligence method for training a learning model for detecting a transportation service related safety issue. The method includes receiving a first sample description of an incident and a known safety issue of the incident. The first sample description includes first set of labeled texts. The method further includes segmenting the first sample description into word segments. The method also includes determining word vectors of the respective word segments and determining features of the first set of labeled texts in the first sample description. The method additionally includes training the learning model based on the word vectors and the features and providing the learning model for automatically detecting the transportation service related safety issue from an incident description.

In a further aspect, embodiments of the disclosure further provide a non-transitory computer-readable medium having instructions stored thereon that, when executed by at least one processor, cause the at least one processor to perform an artificial intelligence method for detecting a transportation service related safety issue. The method includes receiving a first sample description of an incident and a known safety issue of the incident. The first sample description includes first set of labeled texts. The method further includes segmenting the first sample description into word segments. The method also includes determining word vectors of the respective word segments and determining features of the first set of labeled texts in the first sample description. The method additionally includes training the learning model based on the word vectors and the features and providing the learning model for automatically detecting the transportation service related safety issue from an incident description.

It is to be understood that both the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the invention, as claimed.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG. 1 illustrates a schematic diagram of an exemplary transportation service related safety issue detection system, according to embodiments of the disclosure.

FIG. 2 illustrates a schematic diagram of an exemplary system for training a detection model for transportation service related safety issue detection, according to embodiments of the disclosure.

FIG. 3 illustrates a flowchart of an exemplary method for training a detection model for transportation service related safety issue detection, according to embodiments of the disclosure.

FIG. 4 illustrates a schematic diagram of an exemplary system for transportation service related safety issue detection based on a detection model, according to embodiments of the disclosure.

FIG. 5 illustrates a flowchart of an exemplary application of the exemplary method for transportation service related safety issue detection based on a detection model, according to embodiments of the disclosure.

DETAILED DESCRIPTION

Reference will now be made in detail to the exemplary embodiments, examples of which are illustrated in the accompanying drawings. Wherever possible, the same reference numbers will be used throughout the drawings to refer to the same or like parts.

FIG. 1 illustrates a schematic diagram of an exemplary transportation service related safety issue detection system (referred to as “safety issue detection system 100” ) , according to embodiments of the disclosure. Consistent with the present disclosure, safety issue detection system 100 is configured to detect safety issues from incident descriptions (e.g., incident descriptions 103) based on a detection model 105 trained using sample incident descriptions and corresponding known safety issues (e.g., included in training data 101) . In some embodiments, safety issue detection system 100 may include components shown in FIG. 1, including a training database 110, a model training device 120, a detection device 130, an incident description dataset 140, a terminal device 150, a network 160 and a customer service 170 to facilitate communications among the various components. It is contemplated that safety issue detection system 100 may include more or less components compared to those shown in FIG. 1.

Consistent with the present disclosure, safety issue detection system 100 may receive incident descriptions (e.g., sample passenger incident descriptions or customer service incident descriptions as part of training data 101 or incident descriptions 103) from terminal device 150 or customer service 170. For example, terminal device 150 may be a mobile phone, a desktop computer, a laptop, a PDA, a robot, a kiosk, etc. or any in-vehicle device that can record the passenger incident description. Terminal device 150 may include a user interaction interface configured to receive the passenger incident descriptions provided by one or more passengers. In some embodiments, terminal device 150 may include a microphone and a voice recording device such as an analog recording device (e.g., phonograph record or magnetic tape recording) or a digital recording device (e.g., a digital recorder) , for recording the incident description provided by the passenger orally. Terminal device 150 may additionally or alternatively include a transcribe module for transcribing the recorded audio records into texts (e.g., incident descriptions) . In some embodiments, terminal device 150 may record the incident descriptions as texts. For example, terminal device 150 may include a keyboard, a stylus, or a touch screen for receiving text inputs. In some embodiment, similar to terminal device 150, customer service 170 may also be equipped with a microphone and a recording device for recording the incident description provided by the passenger over the phone and may also have a transcribe module for transcribing the recorded audio data into incident descriptions in text.

In some embodiments, the same incident may be described by passenger incident descriptions recorded by in-vehicle device during a sample transportation service, and customer service incident descriptions recorded during a phone conversation between the passenger using the sample transportation service and customer service 170. These incident descriptions are referred to as “sample” incident descriptions consistent with the present disclosure. They may be collected and used as part of training data 101 for safety issue detection system 100 to train detection model 105. For example, sample incident descriptions and known safety issues of the respective service trip, during which the incident happened as described in the sample incident descriptions may be used by model training device 120 to train detection model 105. The known incidents of the service trip may be benchmark detections made by customer service operators based on the sample incident descriptions. Sample incident descriptions and their respective known safety issues may be stored in pairs in training database 110 as training data 101.

In some embodiments, based on trained detection model 105, safety issue detection system 100 may process and make automatic detections from incident descriptions 103. For example, detection device 130 may detect a safety issue of the service trip during which passenger made incident description 103, or during which the phone conversation between the passenger using the transportation service and customer service 170 is recorded. Incident descriptions 103 may be stored in incident description database 140. In some embodiments, incident description 103, along with its detection results (e.g., detected safety issues) , may be periodically provided to update training database 110.

As shown in FIG. 1, safety issue detection system 100 may include components for performing two stages, a training stage and a detection stage. To perform the training stage, safety issue detection system 100 may include training database 110 and model training device 120. To perform the detection stage, safety issue detection system 100 may include detection device 130 and incident description database 140. In some embodiments, when a learning model (e.g., detection model 105) for safety issue detections is pre-trained, safety issue detection system 100 may only include detection device 130 and incident description database 140 to perform safety detection related functions.

Safety issue detection system 100 may optionally include network 160 to facilitate the communication among the various components of safety issue detection system 100, such as

database

110 and 140,

devices

120 and 130, and

terminal

150 and 170. For example, network 160 may be a local area network (LAN) , a wireless network, a cloud computing environment (e.g., software as a service, platform as a service, infrastructure as a service) , a client-server, a wide area network (WAN) , etc. In some embodiments, network 160 may be replaced by wired data communication systems or devices.

In some embodiments, the various components of safety issue detection system 100 may be remote from each other or in different locations and be connected through network 160 as shown in FIG. 1. In some alternative embodiments, certain components of safety issue detection system 100 may be located on the same site or inside one device. For example, training database 110 may be located on-site with or be part of model training device 120. As another example, model training device 120 and detection device 130 may be inside the same computer or processing device, such as inside customer service 170.

As shown in FIG. 1, model training device 120 may communicate with training database 110 to receive one or more sets of training data 101. Each set of training data 101 may include a sample incident description (e.g., a passenger incident description and/or a customer service incident description) and its corresponding ground truth safety issues detection that indicates the known safety issue of the service trip. Model training device 120 may use training data 101 received from training database 110 to train a learning model, e.g., detection model 105, for detecting safety issues for a service trip based on the passenger’s descriptions or the phone conversations between the passenger and customer service 170. Model training device 120 may be implemented with hardware specially programmed by software that performs the training process. For example, model training device 120 may include a processor and a non-transitory computer-readable medium (discussed in detail in connection with FIG. 2) . The processor may conduct the training by performing instructions of a training process stored in the computer-readable medium. Model training device 120 may additionally include input and output interfaces to communicate with training database 110, network 160, and/or a user interface (not shown) . The user interface may be used for selecting sets of training data, adjusting one or more parameters of the training process, selecting or modifying a framework of the learning model, and/or manually or semi-automatically providing detection results associated with a sample incident description for training.

Consistent with the present disclosure, model training device 120 may pre-process the sample descriptions before training detection model 105. The pre-processing may use a word embedding model. For example, each sample incident description may be segmented into word segments and the word segments may be provided to the word embedding model as input. The word embedding model may assign a word vector to each word segment. In some embodiments, the pre-processing may further include text labeling. The text is labeled according to a key-word list for feature extraction. A key-word list includes words or phrases manually generated and/or updated by operator’s indicative of safety issues. For example, if the text includes words such as “drinking, ” “wired” and “rude, ” which are among the key-word list, those words may be labeled within the text. Model training device 120 may extract features from the labeled text and construct a feature matrix.

Consistent with some embodiments, detection model 105 may further include a convolutional neural network (CNN) model to process data where features determined based on the labeled texts and the word vectors from the word embedding model may be used as input. The architecture of a CNN model includes a stack of distinct layers that transform the input into the output. As used herein, “training” a learning model refers to determining one or more parameters of at least one layer in the learning model. For example, a convolutional layer of a CNN model may include at least one filter or kernel. One or more parameters, such as kernel weights, size, shape, and structure, of the at least one filter may be determined by e.g., a backpropagation-based training process. Consistent with some embodiments, detection model 105 may be trained using supervised learning.

Detection device 130 may receive detection model 105 from model training device 120. Detection device 130 may include a processor and a non-transitory computer-readable medium (not shown) . The processor may perform instructions of a safety issue detection process stored in the medium. Detection device 130 may additionally include input and output interfaces to communicate with incident description database 140, network 160, and/or a user interface (not shown) . The user interface may be used for selecting an incident description 103 for safety issues detection, initiating the detection process, or displaying a detection result 107.

Detection device 130 may communicate with incident description database 140 to receive one or more incident descriptions 103. In some embodiments, the incident descriptions 103 stored in incident description database 140 may be received from terminal device 150. Detection device 130 may use the trained model received from model training device 120 to detect safety issues of a service trip the scenario of which are described by incident description 103, and output detection result 107.

FIG. 2 illustrates a schematic diagram of an exemplary system 200 for transportation service related safety issue detection, according to embodiments of the disclosure. Consistent with the present disclosure, system 200 may be an embodiment of model training device 120. In some embodiments, as shown in FIG. 2, system 200 may include a communication interface 202, a processor 204, a memory 206, and a storage 208. In some embodiments, system 200 may have different modules in a single device, such as an integrated circuit (IC) chip (e.g., implemented as an application-specific integrated circuit (ASIC) or a field-programmable gate array (FPGA) ) , or separate devices with dedicated functions. In some embodiments, one or more components of system 200 may be located in a cloud or may be alternatively in a single location (such as inside a mobile device) or distributed locations. Components of system 200 may be in an integrated device or distributed at different locations but communicate with each other through a network (not shown) . Consistent with the president disclosure, system 200 may be configured to train detection model 105 based on training data 101, which is provided to detection device 130 for processing incident description 103.

Communication interface 202 may send data to and receive data from components such as training database 110 via communication cables, a Wireless Local Area Network (WLAN) , a Wide Area Network (WAN) , wireless networks such as radio waves, a cellular network, and/or a local or short-range wireless network (e.g., Bluetooth ^TM) , or other communication methods. In some embodiments, communication interface 202 may include an integrated services digital network (ISDN) card, cable modem, satellite modem, or a modem to provide a data communication connection. As another example, communication interface 202 may include a local area network (LAN) card to provide a data communication connection to a compatible LAN. Wireless links can also be implemented by communication interface 202. In such an implementation, communication interface 202 can send and receive electrical, electromagnetic or optical signals that carry digital data streams representing various types of information.

Consistent with some embodiments, communication interface 202 may receive training data 101 including sample incident descriptions (e.g., passenger incident descriptions and/or customer service incident descriptions) and their respective known safety issues from training database 110.

The sample incident descriptions may be received as texts or in their original format as acquired by terminal device 150 or customer service 170, such as an audio. A sample incident description may include one sentence or multiple sentences that describe the scenario and/or surrounding of a safety issue. When the incident description is made orally, terminal device 150 or customer service 170 may transcribe the incident description to text data and may transmit the text data to communication interface 202 of system 200. In some other embodiments, communication interface 202 may receive the incident description in its original format, such as an audio, and a transcription model (not shown) in system 200 may transcribe the audio into text data. Communication interface 202 may further provide the text data to memory 206 and/or storage 208 for storage or to processor 204 for processing.

Processor 204 may include any appropriate type of general-purpose or special-purpose microprocessor, digital signal processor, or microcontroller. Processor 204 may be configured as a separate processor module dedicated to training a learning model, e.g., detection model 105. Alternatively, processor 204 may be configured as a shared processor module for performing other functions in addition to model training.

Memory 206 and storage 208 may include any appropriate type of mass storage provided to store any type of information that processor 204 may need to operate. Memory 206 and storage 208 may be a volatile or non-volatile, magnetic, semiconductor, tape, optical, removable, non-removable, or other type of storage device or tangible (i.e., non-transitory) computer-readable medium including, but not limited to, a ROM, a flash memory, a dynamic RAM, and a static RAM. Memory 206 and/or storage 208 may be configured to store one or more computer programs that may be executed by processor 204 to perform functions disclosed herein. For example, memory 206 and/or storage 208 may be configured to store program (s) that may be executed by processor 204 to train detection model 105.

Memory 206 and/or storage 208 may be further configured to store information and data used by processor 204. For instance, storage 208 may be configured to store a knowledge database including the various types of data associated with service trips, scenarios, safety issues, and other safety related data. In some embodiments, knowledge database may include various lists used for automatically recognizing safety issues from incident descriptions, such as a key-word list manually generate and/or updated by operators.

Consistent with the present disclosure, knowledge database may further include words vectors for training the word embedding model. In some embodiments, the word vectors are determined using word embedding, which maps the words to vectors of real numbers. In some embodiments, the word vectors may be of several hundred dimensions.

In some embodiments, memory 206 and/or storage 208 may also store intermediate data such as the word segments in sample incident descriptions, feature maps output by layers of the learning model, and optimization loss functions, etc. Memory 206 and/or storage 208 may additionally store various learning models including their model parameters, such as a CNN model and a word embedding model, etc. that will be described. The various types of data may be stored permanently, removed periodically, or disregarded immediately after the data is processed.

As shown in FIG. 2, processor 204 may include multiple modules, such as a word embedding unit 240, a feature determination unit 242, a training data integration unit 244 and a CNN training unit 246, and the like. These modules (and any corresponding sub-modules or sub-units) can be hardware units (e.g., portions of an integrated circuit) of processor 204 designed for use with other components or software units implemented by processor 204 through executing at least part of a program. The program may be stored on a computer-readable medium, and when executed by processor 204, it may perform one or more functions. Although FIG. 2 shows units 240-246 all within one processor 204, it is contemplated that these units may be distributed among different processors located closely or remotely with each other.

In some embodiments, units 242-246 of FIG. 2 may execute computer instructions to perform the training. For example, FIG. 3 illustrates a flowchart of an exemplary method 300 for training detection model 105, according to embodiments of the disclosure. In some embodiments, before being applied to method 300, sample incident descriptions within training data 101 may be labeled using key-word list to improve the efficiency of the detection process. For example, operators may manually label and/or update the label of the sample incident descriptions based on the key-word list.

Method 300 may include steps S302-S320 as described below. In some embodiments, communication interface 202 may perform steps S302 and S310, word embedding unit 240 may perform steps S304-S306 and S312-S314, feature determination unit 242 may perform steps S308 and S316, training data integration unit 244 may perform step S318, and CNN training unit 246 may perform step S320. It is to be appreciated that some of the steps and units may be optional to perform the disclosure provided herein. For example, S318 and training data integration unit 244 may be optional to perform the method if only one type of incident description (e.g., only the passenger incident descriptions or customer service incident descriptions) is used as training data (e.g., training data 101) . Further, some of the steps may be performed simultaneously, or in a different order than shown in FIG. 2. For example, S310-S316 may be performed at the same time while S302-S308 is being performed.

In step 302, communication interface 202 may receive training data 101 including a first sample description of an incident (e.g., the passenger incident descriptions) and known safety incidents corresponding to the first sample description of the incidents. In some embodiments, the first sample description may be a passenger incident description recorded by an in-vehicle device during a sample transportation service. In some embodiments, a large number of training data may be received to train the learning model. Each sample incident description may include one or more sentences that describe a safety issue. For example, the description may be “The driver was not driving according to the predetermined route as shown in the App! ”

In step S304, word embedding unit 240 may segment each sample description into multiple word segments. A word segment is the smallest unit in a sentence that has semantic meanings. A word segment may be a word or a combination of two or more words. If the incident description includes multiple sentences, it may be segmented into different sentences first. In some embodiments, each sample description may be segmented using a sentence segmentation model trained using sample sentences and known word segments of those sentences. Applying the segmentation model, each sample incident description is segmented into a plurality of word segments.

In step S306, word embedding unit 240 may determine word vectors of the respective word segments using, e.g., a word embedding model.

In step S308, feature determination unit 242 may determine features of a first set of labeled texts in the first incident description. For example, feature determination unit 242 may construct a first feature matrix M ₁ for the labeled texts. Text labeling may be performed manually by operators or automatically by safety issue detection system 100. In some embodiments, the text is labeled according to the key-word list. For example, if the text includes words such as drinking, wired or rude, which are among the key-word list generated by the operators, those words will be labeled within the text.

In some embodiments, method 300 may further include S310-S320 in which a second sample description may be received and may be used to additionally train the learning model. In step S310, a second sample description of the incident, different from the first sample description may be received. In some embodiments, the second sample description may be a customer service incident description recorded during a phone conversation between the passenger using the sample transportation service and customer service 170. In some embodiments, to increase the training efficiency, all the app-related training data may be differentiated from all the customer service related training data. For example, all the app-related training data may be suffixed as “_U” and all the customer service related training data may be suffixed as “_S” .

In step S312, the second sample description of the incident may be segmented by word embedding unit 240 similar to step S304. In step S314, additional word vectors of the respective additional word segments may be determined based on the additional information provided by the second sample description of the incident by word embedding unit 240, similar to step S306. In step S316, additional features of a second set of labeled texts in the second sample description may be determined, similar to step 308.

In step S318, the features of the second set of labeled texts may be combined with the feature of the first set of labeled texts. In some embodiments, training data integration unit 244 may match the first sample description with a second sample description describing the same incident. For example, training data integration unit 244 may match the first sample description with the second sample description with the same incident identification number. Training data integration unit 244 may also combine the feature matrix of the matched first sample description and second sample description.

In step S320, CNN processing unit 246 may train the learning model based on the word vectors and the combined features. For example, CNN processing unit 246 may use the combined feature matrix together with the word vectors as input of CNN to train the learning model. In some embodiments, CNN processing unit 246 may generate a detection result as the output of CNN based on the input. In some embodiments, the learning model may be trained by minimizing a difference between the output obtained from CNN and a feature map corresponding to the known safety issues associated with the incident described by the first sample description and the second sample description. In some embodiments, the difference may be a mean square loss (i.e., norm-2 difference) between the two feature maps. Any suitable method may be used to solve the optimization problem, such as various iterative methods.

In some embodiments, if only the first sample description (e.g., passenger description recorded by the transportation service application) is available as training data, method 300 may skip steps S310-318, and proceed to step S320 directly after S308 to train detection model 105 based on the word vectors and features derived from the first incident description. Similarly, in some embodiments, if only the second sample description (e.g., description recorded by the customer service) is available as training data, method 300 may skip steps S302-308, and include steps S310-S316 and S320.

FIG. 4 illustrates a schematic diagram of an exemplary system 400 for transportation service related safety issue detection based on detection model 105, according to embodiments of the disclosure. Consistent with the present disclosure, system 400 may be an embodiment of detection device 130. In some embodiments, as shown in FIG. 4, system 400 may include a communication interface 402, a processor 404, a memory 406, and a storage 408. In some embodiments, system 400 may have hardware components and configurations similar to system 200. Consistent with the president disclosure, system 400 may be configured to detect safety issues from incident descriptions 103 based on detection model 105 provided by model training device 120.

Communication interface 402 may be configured similarly as communication interface 202. In some embodiments, communication interface 402 may send data to and receive data from components such as model training device 120, incident description database 140, training database 110 and display 450. For example, communication interface 402 may receive detection model 105 from model training device 120, and incident descriptions 103 from incident description database 140. Incident descriptions 103 may be provided by passenger 430 through terminal device 150 or by customer service 170.

Processor 404 may include hardware components similar to those in processor 204. Processor 404 may be configured as a separate processor module dedicated to making safety issues detection using a learning model. Alternatively, processor 404 may be configured as a shared processor module for performing other functions in addition to safety issues detection. Memory 406 and storage 408 may be similar to memory 206 and storage 208. For example, memory 406 and/or storage 408 may be configured to store program (s) that may be executed by processor 404 to detect safety issues based on detection model 105. Storage 408 may also store the knowledge database similar to that is stored in storage 208.

In some embodiments, memory 406 and/or storage 408 may also store intermediate data such as the word segments in incident descriptions, feature maps output by layers of the learning model, etc. Memory 406 and/or storage 408 may additionally store various learning models including their model parameters, such as detection model 105, and the word embedding model, etc.

As shown in FIG. 4, processor 404 may include multiple modules, such as a word embedding unit 440, a feature determination unit 442, a data integration unit 444 and a CNN processing unit 446, and the like. These modules (and any corresponding sub-modules or sub-units) can be hardware units (e.g., portions of an integrated circuit) of processor 404 designed for use with other components or software units implemented by processor 404 through executing at least part of a program. The program may be stored on a computer-readable medium, and when executed by processor 404, it may perform one or more functions. Although FIG. 4 shows units 440-446 all within one processor 404, it is contemplated that these units may be distributed among different processors located closely or remotely with each other.

In some embodiments, units 442-446 may execute computer instructions to perform the training. FIG. 5 illustrates a flowchart of an exemplary method 500 for transportation service related safety issue detection based on detection model 105, according to embodiments of the disclosure. Method 500 may be implemented by system 400 and particularly processor 404 or a separate processor not shown in FIG. 4. Method 500 may include steps S502-S522 as described below. It is to be appreciated that some of the steps may be optional to perform the disclosure provided herein. Further, some of the steps may be performed simultaneously, or in a different order than shown in FIG. 5.

In step 502, communication interface 402 may receive incident descriptions 103 including a first incident description of an incident. In some embodiments, the first incident description may be a passenger incident description recorded by an in-vehicle device during a transportation service. For example, the description may be “The driver was not driving according to the predetermined route as shown in the App! ”

In steps S504-S508, word embedding unit 440 and feature determination unite 442 may pre-process the first incident description, similar to step S304-S308, to obtain word vectors of the word segments and features of labeled texts.

In step S510, a second incident description of the same incident, different from the first incident description may be received. In some embodiments, the second incident description may be a customer service incident description recorded during a phone conversation between the passenger using the transportation service and customer service 170.

In steps S512-S516, word embedding unit 440 and feature determination unite 442 may pre-process the second incident description, similar to step S312-S316, to obtain additional word vectors of the word segments in the second incident description and additional features of labeled texts. In step S518, the features derived from the second incident description may be combined with the features derived from the first incident description, similar to step S318.

In step S520, CNN processing unit 446 may apply detection model 105 to the combined features and word vectors to detect one or more safety issues. In some embodiments, CNN processing unit 446 may input the word vectors and the compared feature matrix to trained detection model 105 to detect the safety issues and provide detection result 107.

In some embodiments, if only the first incident description (e.g., passenger description recorded by the transportation service application in terminal device 150) is available, method 500 may skip steps S510-518, and proceed to step S520 directly after S508. Method 500 may select detection model 105 that is trained using solely descriptions recorded by terminal device 150 and apply the selected model in step S520. Similarly, in some embodiments, if only the second incident description (e.g., description recorded by the customer service) is available, method 500 may skip steps S502-508, and include steps S510-S516 and S520. Accordingly, method 500 may select detection model 105 that is trained using solely descriptions recorded by customer service 170 and apply the selected model in step S520.

In some embodiments, communication interface 402 may also provide detection result 107 output by CNN. For example, as shown by FIG. 4, detection result 107 may be provided to customer service 170 through display 450. Display 450 may include a display such as a Liquid Crystal Display (LCD) , a Light Emitting Diode Display (LED) , a plasma display, or any other type of display, and provide a Graphical User Interface (GUI) presented on the display for user input and data depiction. The display may include a number of different types of materials, such as plastic or glass, and may be touch-sensitive to receive inputs from the user. For example, the display may include a touch-sensitive material that is substantially rigid, such as Gorilla Glass ^TM, or substantially pliable, such as Willow Glass ^TM. In some embodiments, display 450 may be part of customer service 170.

In step S522, the detected safety issue (e.g., detection result 107) and the respective incident descriptions may be added to training database 110 as training data 101. For example, communication interface 402 may update training database 110 in real time, in a periodic manner, in a request-response manner, etc. After training database 110 is updated, model training device 120 may repeat method 300 to train an updated detection model 105.

Another aspect of the disclosure is directed to a non-transitory computer-readable medium storing instruction which, when executed, cause one or more processors to perform the methods, as discussed above. The computer-readable medium may include volatile or non-volatile, magnetic, semiconductor, tape, optical, removable, non-removable, or other types of computer-readable medium or computer-readable storage devices. For example, the computer-readable medium may be the storage device or the memory module having the computer instructions stored thereon, as disclosed. In some embodiments, the computer-readable medium may be a disc or a flash drive having the computer instructions stored thereon.

It will be apparent to those skilled in the art that various modifications and variations can be made to the disclosed system and related methods. Other embodiments will be apparent to those skilled in the art from consideration of the specification and practice of the disclosed system and related methods.

It is intended that the specification and examples be considered as exemplary only, with a true scope being indicated by the following claims and their equivalents.

Claims

An artificial intelligence system for training a learning model for detecting a transportation service related safety issue, comprising:

a storage device configured to store a first sample description of an incident and a known safety issue of the incident, wherein the first sample description includes first set of labeled texts;

at least one processor configured to:

segment the first sample description into word segments;

determine word vectors of the respective word segments;

determine features of the first set of labeled texts in the first sample description; and

train the learning model based on the word vectors and the features; and

a communication interface configured to provide the learning model for automatically detecting the transportation service related safety issue from an incident description.
The artificial intelligence system of claim 1, wherein the first sample description is a passenger incident description recorded by an in-vehicle device during a sample transportation service.
The artificial intelligence system of claim 1, wherein the first sample description is

a customer service incident description recorded during a phone conversation between

a passenger using a sample transportation service and a customer service.
The artificial intelligence system of claim 1, wherein the storage device is further configured to store a second sample description of the incident, different from the first sample incident description,

wherein the at least one processor is further configured to:

segment the second sample description into additional word segments;

determine additional word vectors of the respective additional word segments; and

train the learning model additionally using the additional word vectors.
The artificial intelligence system of claim 4, wherein the second sample description includes a second set of labeled texts, wherein the at least one processor is further configured to:

associate the first set of labeled texts with the second set of labeled texts;

determine additional features of the second set of labeled texts in the first sample description;

combine the features of the first set of labeled texts and the additional features of the second set of labeled texts according to the association; and

train the learning model additionally using the combined features.
The artificial intelligence system of claim 5, wherein the first sample description is a passenger incident description recorded by an in-vehicle device during a sample transportation service, and the second sample description is a customer service incident description recorded during a phone conversation between the passenger using the sample transportation service and a customer service.
The artificial intelligence system of claim 1, wherein the learning model is a convolutional neural network (CNN) .
The artificial intelligence system of claim 1, wherein the at least one processor is further configured to:

segment the incident description into word segments;

determine word vectors of the respective word segments;

receive labels on a subset of texts in the incident description;

determine features of the labeled texts in the incident description; and

detect the transportation service related safety issue based on the word vectors and features using the trained learning network.
The artificial intelligence system of claim 8, wherein the at least one processor is further configured to:

receive an audio recording recorded during a transportation service; and

transcribe the audio recording into the incident description.
The artificial intelligence system of claim 9, wherein the audio recording is an in-vehicle recording recorded during the transportation service, or a customer service recording recorded during a phone conversation between a passenger using the transportation service and a customer service center.
An artificial intelligence method for training a learning model for detecting a transportation service related safety issue, comprising:

receiving, by a communication interface, a first sample description of an incident and a known safety issue of the incident, wherein the first sample description includes first set of labeled texts;

segmenting, by at least one processor, the first sample description into word segments;

determining, by the at least one processor, word vectors of the respective word segments;

determining, by the at least one processor, features of the first set of labeled texts in the first sample description; and

training, by the at least one processor, the learning model based on the word vectors and the features; and

providing, by the communication interface, the learning model for automatically detecting the transportation service related safety issue from an incident description.
The artificial intelligence method of claim 11, wherein the first sample description is a passenger incident description recorded by an in-vehicle device during a sample transportation service.
The artificial intelligence method of claim 11, wherein the first sample description is

a customer service incident description recorded during a phone conversation between

a passenger using a sample transportation service and a customer service.
The artificial intelligence method of claim 11, further comprising:

receiving a second sample description of the incident, different from the first sample incident description,

segmenting the second sample description into additional word segments;

determining additional word vectors of the respective additional word segments; and

training the learning model additionally using the additional word vectors.
The artificial intelligence method of claim 11, wherein the second sample description includes a second set of labeled texts, wherein the method further comprises:

associating the first set of labeled texts with the second set of labeled texts;

determining additional features of the second set of labeled texts in the first sample description;

combining the features of the first set of labeled texts and the additional features of the second set of labeled texts according to the association; and

training the learning model additionally using the combined features.
The artificial intelligence method of claim 11, wherein the first sample description is a passenger incident description recorded by an in-vehicle device during a sample transportation service, and the second sample description is a customer service incident description recorded during a phone conversation between the passenger using the sample transportation service and a customer service.
The artificial intelligence method of claim 11, wherein the learning model is a convolutional neural network (CNN) .
The artificial intelligence method of claim 11, wherein the word vectors are determined using a word embedding model.
The artificial intelligence method of claim 11, further comprising:

receiving an audio recording recorded during a transportation service;

transcribing the audio recording into the incident description;

segmenting the incident description into word segments;

determining word vectors of the respective word segments;

receiving labels on a subset of texts in the incident description;

determining features of the labeled texts in the incident description; and

detecting the transportation service related safety issue based on the word vectors and features using the trained learning network.
A non-transitory computer-readable medium having instructions stored thereon that, when executed by at least one processor, cause the at least one processor to perform an artificial intelligence method for training a learning model for detecting a transportation service related safety issue, the artificial intelligence method comprising:

receiving a first sample description of an incident and a known safety issue of the incident, wherein the first sample description includes first set of labeled texts;

segmenting the first sample description into word segments;

determining word vectors of the respective word segments;

determining features of the first set of labeled texts in the first sample description; and

training the learning model based on the word vectors and the features; and

providing the learning model for automatically detecting the transportation service related safety issue from an incident description.