WO2018120963A1 - 一种基于反馈的自适应主客观权重上下文感知系统及其工作方法 - Google Patents

一种基于反馈的自适应主客观权重上下文感知系统及其工作方法 Download PDF

Info

Publication number
WO2018120963A1
WO2018120963A1 PCT/CN2017/104988 CN2017104988W WO2018120963A1 WO 2018120963 A1 WO2018120963 A1 WO 2018120963A1 CN 2017104988 W CN2017104988 W CN 2017104988W WO 2018120963 A1 WO2018120963 A1 WO 2018120963A1
Authority
WO
WIPO (PCT)
Prior art keywords
context information
context
subjective
information
user
Prior art date
Application number
PCT/CN2017/104988
Other languages
English (en)
French (fr)
Inventor
许宏吉
周英明
房海腾
潘玲玲
孙君凤
许征征
杜保臻
Original Assignee
山东大学
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 山东大学 filed Critical 山东大学
Priority to JP2019532041A priority Critical patent/JP6771105B2/ja
Publication of WO2018120963A1 publication Critical patent/WO2018120963A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N5/00—Computing arrangements using knowledge-based models
    • G06N5/01—Dynamic search techniques; Heuristics; Dynamic trees; Branch-and-bound
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N5/00—Computing arrangements using knowledge-based models
    • G06N5/04—Inference or reasoning models

Definitions

  • the invention relates to a feedback-based adaptive subjective and objective weight context awareness system and a working method thereof, and belongs to the technical field of context awareness.
  • the context-aware system implements a device-centric to human-centric transformation of the computing system.
  • the computing system can automatically collect context information of interest and perceive changes in the application context, and actively provide relevant applications to the user based on the collected context information. service.
  • the wireless sensor network technology is becoming more and more mature, and a wireless sensor network composed of a large number of microprocessor nodes with computing power can be used to obtain various information anytime and anywhere.
  • the context-aware system is also capable of making efficient, intelligent, and personalized services based on massive amounts of information.
  • context information may reflect the situation or state of the same event, but have different influence on the events perceived by the system, and since the ultimate purpose of the context-aware system is to provide users with "people-oriented" personalized services, This requires the system to reflect the user's personal preferences in the process of fusion reasoning of context information, reflecting that the various context information has different influences on the user in the process of system inference and fusion, that is, the various context information is in the process of inference and fusion. Have different weights. Therefore, how to reflect the difference of different types of context information in reasoning fusion and the degree of influence on decision-making, and how to dynamically adjust the system according to user feedback to improve the perception accuracy becomes a major challenge of context-aware technology.
  • the existing context-aware systems are not perfect, some are less intelligent, and some are more restricted by the application domain.
  • how to more accurately and intelligently provide users with personalized Services are still the focus of research in the context of contextual awareness.
  • the present invention provides a feedback-based adaptive subjective and objective weight context aware system.
  • the present invention also provides a method of working with the context aware system described above.
  • the context-aware system will get the final decision information after processing a large amount of different types of original context information.
  • different types of context information can reflect the situation or state of the same event, but have different influence on the events perceived by the system.
  • the system should fully consider the subjective preference information of users and achieve "people-oriented".
  • the adaptive subjective and objective weighted context-aware system based on feedback considers both the user's preference information and the objective influence of various context information on decision-making. Selecting appropriate context information types for different perceptual events, in the context information fusion reasoning process Different weight factors are assigned to different kinds of context information.
  • the mechanism dynamically updates and optimizes the contribution rate of subjective weight and objective weight of each type of context information based on the feedback information of the user, and then obtains the subjective and objective weights of various context information, that is, the optimized combination value of subjective weight and objective weight, and improves
  • the intelligence and accuracy of the context-aware system can provide users with more personalized services.
  • the original context information is diverse, and heterogeneous data inevitably exists.
  • the similar context information has certain similarities, and it is easier to reduce and merge. Therefore, the fusion of similar context information is first carried out, and then the dynamic subjective and objective weight information is combined to carry out the fusion of multiple heterogeneous context information, which can be more efficiently obtained. Accurate perception of results.
  • a feedback-based adaptive subjective and objective weight context awareness system includes a raw context information collection module, a context fusion module, a context inference module, a context application module, and a user module.
  • the original context information collection module is connected to the context fusion module, and the context fusion module, the context inference module, the context application module, and the user module are cyclically connected in sequence;
  • the original context collection module is configured to collect original context information from different information sources in different manners, where the original context information refers to raw data collected from different sensors; the context fusion module is configured to perform original context information. a fusion process to extract primary context information usable by the context inference module; the context inference module is configured to infer various different primary context information to derive an advanced context that can be directly used by the context application module
  • the context application module utilizes the advanced context information to adjust a corresponding application or device to provide an appropriate service for the user; the user module extracts explicit or implicit feedback from the user, and after quantitative evaluation, converts to each class
  • the judgment information of the accuracy of the context information is sent to the subjective objective weight management unit for optimizing the respective contribution rates of subjective weights and objective weights for each type of context information; the explicit feedback refers to feedbacks actively generated by the user. Satisfaction scoring mechanism Set to user data, the implicit feedback system refers to a user context-aware expression recognition, user behavior analysis indirectly inferred user data.
  • the context fusion module includes a context information pre-processing unit and a host guest connected in sequence View weight management unit, context information fusion unit;
  • the context information pre-processing unit performs data modeling on the original context information, and the modeling mode is “perceive type + perceptual information”, and classifies the original context information by means of an averaging method, a least square method, a maximum likelihood estimation method or
  • the Kalman filtering method performs missing value processing and fusion operations on the same context information;
  • the subjective objective weight management unit is configured to: identify, according to a system-aware event type, a type of context information that needs to be assigned a weight, and calculate a subjective weight of each type of context information by using a subjective weighting algorithm, where the subjective weighting algorithm includes: a Delphi algorithm, Cyclic scoring method, binomial coefficient method and Analytic Hierarchy Process (AHP); then objective weighting method is used to calculate the objective weight of various context information, including Principal Components Analysis , PCA), coefficient of variation method, entropy weight method and multi-objective programming method, dynamically calculate the subjective weight of each type of context information and the contribution rate of objective weight according to the feedback information of user feedback, and finally obtain various kinds of context information according to the contribution rate.
  • a subjective weighting algorithm includes: a Delphi algorithm, Cyclic scoring method, binomial coefficient method and Analytic Hierarchy Process (AHP); then objective weighting method is used to calculate the objective weight of various context information, including Principal Components Analysis , PCA), coefficient of variation method,
  • the context information fusion unit combines multiple types of context information and subjective and objective weights given by the subjective objective weight management unit with a plurality of information fusion algorithms to obtain primary context information, and the information fusion algorithm includes neural network based fusion. Algorithm, fusion algorithm based on Kalman filter, fusion algorithm based on fuzzy theory and fusion algorithm based on DS evidence theory.
  • the context inference module includes a decision management unit and a context information inference unit that are sequentially connected;
  • the decision management unit selects an appropriate reasoning method according to the type of the primary context information and the type of the perceived event, and assigns different weights to the results of the plurality of information fusion algorithms to improve the inference accuracy;
  • the context information inference unit infers the primary context information generated by the different fusion algorithms of the context information fusion unit according to the inference method given by the decision management unit, and obtains an advanced level that can be directly used by the context application module.
  • Context information includes an ontology reasoning method, a rule-based reasoning method, an evidence-based reasoning method, and a Bayesian network-based reasoning method.
  • the working method of the above context aware system includes the following steps:
  • the context information pre-processing unit performs data modeling on the original context information, and the modeling mode is “perceive type + perceptual information”;
  • the context pre-processing unit classifies the original context information, that is, the context information indicating the same attribute is classified into the original original context information, and the original original context information is passed through the mean method, the least squares method, the maximum likelihood estimation method, or Karl.
  • the Manchester filtering method performs missing value processing and fusion operations;
  • the subjective and objective weight management unit identifies the type of context information that needs to be assigned a weight according to the type of the system-aware event and the correlation between the various types of context information and the event. For example, in the detection of water pollution status, water temperature, pH, etc. are usually obtained. There are dozens of types of context information, but only a few types of key context information such as PH value and conductivity are related to the detection of water pollution status. Therefore, it is only necessary to assign weights to these types of key context information. The total number of types of context information that need to be assigned weights is set to n;
  • the subjective objective weight management unit uses a subjective weighting algorithm to assign weights to n types of context information that need to be assigned weights according to user preferences, and the weight vector is expressed as Ws i represents the subjective weight value of the i-th type of context information;
  • the subjective objective weight management unit uses an objective weighting algorithm to assign weights to n types of context information that need to be assigned weights, and the weight vector is expressed as Wo i represents the objective weight value of the i-th type of context information;
  • the subjective objective weight management unit determines whether there is user feedback from the user module, if yes, step S08 is performed, if not, step S09 is performed;
  • the subjective objective weight management unit calculates the contribution rate of the subjective weight and the objective weight according to the evaluated quantitative information fed back by the user, the contribution rate of the subjective weight is ⁇ , the contribution rate of the objective weight is 1- ⁇ , and the calculation formula is as shown in formula (I). Shown as follows:
  • T i is the total number of feedbacks generated by the user
  • R i is: the number of the i-type context information that is obtained by fusion inference and the result of the user feedback is consistent with the user feedback in the case of user feedback.
  • the initial value of ⁇ is set to 0.5 That is, the subjective weight of each type of context information and the contribution rate of the objective weight are the same;
  • the subjective objective weight management unit synthesizes the subjective weight and the objective weight of each type of context information according to the contribution rate, and obtains the subjective and objective weight values of each type of context information, and the synthesis formula is as shown in formula (II):
  • W i is the subjective and objective weight value of the i-th type of context information
  • the context information fusion unit combines subjective and objective weights, and uses different context information fusion algorithms to respectively fuse multiple types of context information, and each information fusion algorithm obtains a probability vector, which is used to indicate the possibility of all perceptual results of the system. These same or different probability vectors are used as primary context information for context information reasoning;
  • the context inference module uses the inference algorithm given by the decision management unit to infer the primary context information obtained in step S10, and several identical or different probability vectors are inferred by corresponding inference algorithms to obtain a final perceptual result, that is, Advanced context information that is available to the context application module application;
  • the context application module After receiving the advanced context information, the context application module adjusts the corresponding application or device to provide an appropriate service for the user;
  • the user module records the adjustment of the system by the user during the use of the context application or the behavior of the user during the use of the system as explicit feedback or implicit feedback of the user;
  • the user module extracts the user explicit feedback or the implicit feedback, and converts the judgment information into the accuracy of each type of context information, and sends the information to the subjective objective weight management unit to optimize the respective contribution rates of the subjective weight and the objective weight.
  • the feedback-based adaptive subject-objective weight context-aware system of the present invention considers the user's preference information and the objective influence of various context information on the decision, and selects a reasonable context information type for different perceived events in the context.
  • different weighting factors are assigned to different kinds of context information, which improves the efficiency and accuracy of context information fusion reasoning, and can provide users with smarter and personalized services.
  • the feedback-based adaptive subject-objective weight context-aware system dynamically updates and optimizes the contribution rates of subjective weights and objective weights of various context information, and then obtains subjective and objective information of various context information.
  • the weight that is, the optimized combination of subjective weight and objective weight, improves the intelligence and accuracy of the context-aware system, and can provide users with more personalized services.
  • FIG. 1 is a structural framework diagram of a feedback-based adaptive subjective and objective weight context aware system according to the present invention
  • FIG. 2 is a flowchart of the work of the feedback-based adaptive subjective and objective weight context aware system according to the present invention.
  • a feedback-based adaptive subjective and objective weight context awareness system includes an original context information collection module, a context fusion module, a context inference module, a context application module, and a user module.
  • the original context information collection module is connected to the context fusion module, and the context fusion module, the context inference module, the context application module, and the user module are sequentially connected in a loop;
  • the original context collection module is configured to collect original context information from different information sources in different manners.
  • the original context information refers to the original data collected from different sensors;
  • the context fusion module is used to fuse the original context information, and extract the Primary context information for use by the context inference module;
  • the context inference module is configured to infer various different primary context information to derive high level context information that is directly usable by the context application module;
  • the context application module utilizes the Advanced context information, adjusting the corresponding application or device to provide appropriate services for the user;
  • the user module extracts explicit or implicit feedback from the user, and after quantitative evaluation, converts the judgment information into the accuracy of each type of context information, and sends the information to the
  • the subjective and objective weight management unit is used to optimize the contribution rate of the subjective weight and the objective weight of each type of context information;
  • the explicit feedback refers to the feedback behavior actively initiated by the user, such as the user data collected by the satisfaction scoring mechanism.
  • Implicit feedback is Expression Recognition by the user context-aware user behavior analysis system
  • the context fusion module includes a context information preprocessing unit, a subjective objective weight management unit, and a context information fusion unit connected in sequence;
  • the context information preprocessing unit performs data modeling on the original context information, and the modeling mode is “perceive type + perceptual information”, and classifies the original context information by mean method, least square method, maximum likelihood estimation method or Kalman
  • the filtering method performs missing value processing and fusion operation on the same context information
  • the subjective objective weight management unit is configured to: according to the type of system-aware event, confirm the type of context information that needs to be assigned a weight, and calculate a subjective weight of each type of context information by using a subjective weighting algorithm, the subjective weighting algorithm includes: Delphi algorithm, loop scoring Method, Binomial Coefficient Method and Analytic Hierarchy Process (AHP); then objective weighting method is used to calculate the objective weight of various contextual information, including Principal Components Analysis (PCA) ), coefficient of variation method, entropy weight method and multi-objective programming method, dynamically calculate the subjective weight of each type of context information and the contribution rate of objective weight according to the feedback information of user feedback, and finally derive the main information of various context information according to the contribution rate.
  • PCA Principal Components Analysis
  • the context information fusion unit combines the multi-class context information and the subjective objective weights given by the subjective objective weight management unit with a plurality of information fusion algorithms to obtain primary context information, and the information fusion algorithm includes a neural network-based fusion algorithm. Fusion algorithm based on Kalman filter, fusion algorithm based on fuzzy theory and fusion algorithm based on DS evidence theory.
  • the context inference module includes a decision management unit and a context information inference unit that are sequentially connected;
  • the decision management unit selects an appropriate reasoning method according to the type of the primary context information and the type of the perceived event, and assigns different weights to the results of the plurality of information fusion algorithms to improve the inference accuracy;
  • the context information inference unit infers the primary context information generated by the different fusion algorithms of the context information fusion unit according to the inference method given by the decision management unit, and obtains advanced context information that can be directly used by the context application module.
  • the context reasoning method includes an ontology reasoning method, a rule-based reasoning method, an evidence-based reasoning method, and a Bayesian network-based reasoning method.
  • a scene of context-aware computing a smart fitness system
  • users are separately collected by different sensors.
  • Contextual information such as body weight, height, heart rate, and oxygen saturation are represented as I w , I h , I r , and I o , respectively .
  • the user is finally given a fitness program suggestion of appropriate exercise intensity.
  • the last recommendations are three: weaker strength projects, medium strength projects, and higher strength projects. The steps are as follows:
  • the original context information obtained has 1 set of weight information I w , 2 sets of height information I h1 , I h2 , 2 sets of heart rate information I r1 , I r2 , and 1 set of blood oxygen saturation information I o .
  • a set of data I h is obtained , and the two sets of heart rate information I r1 and I r2 are processed to obtain a set of data I r , where no data is missing and the amount of data is small.
  • the fusion algorithm selects the mean method.
  • the perceived event type provides the user with a reasonable exercise intensity fitness program, confirms that the height information is ignored, and the type of context information that needs to be assigned is weight, heart rate, blood oxygen saturation, and the total number of context information types is 3.
  • the weight vector of the three types of context information I w , I r , I o is given by the subjective weighting algorithm.
  • the objective weighting vector is used to assign the objective weight vector of the three types of context information I w , I r , I o to
  • step S08 It is judged whether there is feedback information of the user, that is, whether the user's recommended motion mode for the system is suitable for his/her feedback information. If the feedback information is received, step S08 is performed, if no feedback information is received, step S09 is performed;
  • the subjective weight vector of the three types of context information I w , I r , I o is The objective weight vector is Combine the two weights with formula (II), the formula is
  • the three kinds of context information I w , I r , I o are fused, and each algorithm obtains a suggestion expressed by a probability vector.
  • the fusion result of the first algorithm is suitable for the user with weak strength items and medium-intensity items.
  • the probability of each of the higher-intensity items is (0.6, 0.3, 0.1), and a total of four identical or different probability vectors are obtained.
  • step S10 The four probability vectors obtained in step S10 are inferred by the inference algorithm given by the decision management unit, and finally a final suggestion is given, for example, the user is suitable for a medium-strength fitness program;
  • the system according to the suggestion given in step S11 - the user is suitable for a medium-strength fitness program, automatically displays the type of medium-intensity sports items and related precautions on the screen.
  • the user feedback information is considered to be consistent with the reasoning result of the context-aware system.
  • the user module dynamically saves the number of feedbacks of the user and the number of inference results consistent with the user feedback, and these The information is sent to the subjective and objective weight management unit in real time to dynamically update the contribution rate of the subjective weight and the objective weight.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Computing Systems (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Physics & Mathematics (AREA)
  • Computational Linguistics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Artificial Intelligence (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
  • Stored Programmes (AREA)
  • User Interface Of Digital Computer (AREA)

Abstract

一种基于反馈的自适应主客观权重上下文感知系统及其工作方法。所述上下文感知系统，包括原始上下文信息采集模块、上下文融合模块、上下文推理模块、上下文应用模块以及用户模块，所述用户模块还与上下文融合模块相连。该上下文感知系统在上下文融合模块引入了基于用户反馈的自适应主客观权重分配机制，可以动态调整优化各类上下文信息的主观权重和客观权重各自的贡献率，提高上下文信息融合的准确率和效率，使系统可以为用户提供更智能化、个性化的服务。

Description

一种基于反馈的自适应主客观权重上下文感知系统及其工作方法 技术领域
本发明涉及一种基于反馈的自适应主客观权重上下文感知系统及其工作方法，属于上下文感知的技术领域。
背景技术
随着无线传感器网络、人机交互和智能计算技术的大规模应用，以为用户提供“透明”交互为目的的上下文感知技术获得了迅速的发展。上下文感知系统实现了计算系统从以设备为中心到以人为中心的转变，该计算系统可以自动收集感兴趣的上下文信息并感知应用情境的变化，并依据搜集到的上下文信息主动为用户提供相关应用服务。
无线传感器网络技术日趋成熟，使大量具有计算能力的微处理器节点组成的无线传感网络可以被运用，随时随地获取各种信息。上下文感知系统除了要准确大量的采集与当前应用相关的上下文信息之外，还要能够根据海量的信息，针对用户需求做出高效、智能和个性化的服务。
通常，不同类型的上下文信息可以反映同一事件的情况或状态，但是对系统所感知的事件会有不同的影响力，又由于上下文感知系统的最终目的是为用户提供“以人为本”的个性化服务，这就要求系统能够在上下文信息的融合推理过程中体现用户的个人偏好，反映出对用户而言各类上下文信息在系统推理融合过程中具有不同的影响力，即各类上下文信息在推理融合过程中具有不同的权重。因此，如何体现不同类型上下文信息在推理融合中的差异性和对决策的影响程度，以及如何根据用户反馈动态地对系统进行调整以提高感知准确率就成为上下文感知技术的一大挑战。
此外，近年来，研究者在上下文获取、上下文处理、上下文分发及上下文应用等领域开展了大量的研究，但是很多研究工作都是基于特定应用场景进行的，系统架构与应用逻辑紧密耦合在一起，阻碍了多系统之间的互联及互操作，不利于系统的扩展和复用。
综上所述，现有的上下文感知系统还不是很完善，有些智能性较差，有些受应用领域的限制较强，在改善现存问题的基础上，如何更加精确、智能地为用户提供个性化服务仍然是上下文感知领域的研究重点。
发明内容
针对现有技术的不足，本发明提供一种基于反馈的自适应主客观权重上下文感知系统。
本发明还提供了上述上下文感知系统的工作方法。
上下文感知系统在对海量不同类型的原始上下文信息进行处理后才会得到最终的决策信息。通常情况下，不同类型的上下文信息可以反映同一事件的情况或状态，但是对系统所感知的事件会有不同的影响力。并且，系统在为用户提供服务时，要充分考虑用户的主观偏好信息，做到“以人为本”。基于反馈的自适应主客观权重上下文感知系统既考虑到用户的偏好信息，又考虑到了各类上下文信息对决策的客观影响，针对不同的感知事件选择合理的上下文信息种类，在上下文信息融合推理过程中为不同种类的上下文信息分配不同的权重因子。同时，该机制结合用户的反馈信息动态更新优化各类上下文信息的主观权重和客观权重各自的贡献率，进而得到各类上下文信息的主客观权重，即主观权重和客观权重的优化组合值，提高了上下文感知系统的智能性和精确度，可以为用户提供更加个性化的服务。
原始上下文信息多种多样，不可避免地存在异构数据。同类上下文信息间具有一定的相似性，更容易进行约减和融合，所以先进行同类上下文信息的融合，然后结合动态的主客观权重信息进行多种异类上下文信息的融合，可以更高效地得出精确的感知结果。
本发明的技术方案为：
一种基于反馈的自适应主客观权重上下文感知系统，包括原始上下文信息采集模块、上下文融合模块、上下文推理模块、上下文应用模块以及用户模块。
所述原始上下文信息采集模块与所述上下文融合模块相连，所述上下文融合模块、所述上下文推理模块、所述上下文应用模块以及所述用户模块依次循环连接；
所述原始上下文采集模块用于采取不同的方式从不同的信息源采集原始上下文信息，所述原始上下文信息是指从不同传感器采集到的原始数据；所述上下文融合模块用于对原始上下文信息进行融合处理，提取出可供所述上下文推理模块使用的初级上下文信息；所述上下文推理模块用于对各种不同的初级上下文信息进行推理，得出可供所述上下文应用模块直接使用的高级上下文信息；所述上下文应用模块利用所述高级上下文信息，调整相应应用程序或设备，为用户提供恰当的服务；所述用户模块提取用户显式或隐式反馈，经过量化评估后转化为对每类上下文信息准确性的判断信息，发送至所述主客观权重管理单元，用于对各类上下文信息进行主观权重和客观权重各自贡献率的优化；所述显式反馈是指用户主动做出的反馈行为，如采用满意度评分机制收集到的用户数据，所述隐式反馈是指上下文感知系统通过用户表情识别，用户行为分析间接推断出来的用户数据。
根据本发明优选的，所述上下文融合模块包括依次连接的上下文信息预处理单元、主客 观权重管理单元、上下文信息融合单元；
所述上下文信息预处理单元对原始上下文信息进行数据建模，建模模式为“感知类型+感知信息”，并对原始上下文信息分类，通过均值方法、最小二乘方法、最大似然估计方法或卡尔曼滤波方法对同类上下文信息间进行缺失值处理和融合操作；
所述主客观权重管理单元用于：根据系统感知事件类型，确认需要分配权重的上下文信息种类，并用主观赋权算法计算各类上下文信息的主观权重，所述主观赋权算法包括：Delphi算法、循环打分法、二项系数法和层次分析法(Analytic Hierarchy Process，AHP)；然后用客观赋权法计算各类上下文信息的客观权重，所述客观赋权法包括主成分分析法(Principal Components Analysis，PCA)、变异系数法、熵权法和多目标规划法，根据用户反馈的评估量化信息动态计算各类上下文信息的主观权重和客观权重的贡献率，最后根据贡献率得出各类上下文信息的主客观权重；
所述上下文信息融合单元对所述主客观权重管理单元给出的多类上下文信息和主客观权重用多种信息融合算法进行融合，得到初级上下文信息，所述信息融合算法包括基于神经网络的融合算法，基于卡尔曼滤波的融合算法、基于模糊理论的融合算法和基于D-S证据理论的融合算法。
根据本发明优选的，所述上下文推理模块包括依次连接的决策管理单元和上下文信息推理单元；
所述决策管理单元根据初级上下文信息的种类和感知事件类型选择合适的推理方法，并为多种信息融合算法的结果分配不同的权重以提高推理准确度；
所述上下文信息推理单元根据所述决策管理单元给出的推理方法，对所述上下文信息融合单元的不同融合算法产生的初级上下文信息进行推理处理，得到可供所述上下文应用模块直接使用的高级上下文信息，所述上下文推理方法包括本体推理方法、基于规则的推理方法、基于证据论的推理方法和基于贝叶斯网络的推理方法。
上述上下文感知系统的工作方法，包括步骤如下：
S01：原始上下文信息采集
从各类传感器获取原始上下文信息；
S02：上下文信息建模
所述上下文信息预处理单元对原始上下文信息进行数据建模，建模模式为“感知类型+感知信息”；
S03：上下文信息分类融合
所述上下文预处理单元对原始上下文信息进行分类，即表示同种属性的上下文信息归为同类原始上下文信息，同类原始上下文信息间，通过均值方法、最小二乘方法、最大似然估计方法或卡尔曼滤波方法进行缺失值处理和融合操作；
S04：确认感知所需上下文信息类型
所述主客观权重管理单元根据系统感知事件类型及各类上下文信息与该事件的相关性来确认需要分配权重的上下文信息种类，例如：在水质污染状况检测中，通常会获取到水温、酸碱度等几十类上下文信息，但与水质污染状况检测相关性较高的只有PH值、电导率等几类关键上下文信息，所以只需要对这几类关键上下文信息分配权重。需要分配权重的上下文信息种类数量总数设为n；
S05：计算各类上下文信息主观权重
所述主客观权重管理单元运用主观赋权算法，根据用户偏好，为需要分配权重的n类上下文信息分配权重，权重向量表示为
Figure PCTCN2017104988-appb-000001
Wsi表示第i类上下文信息的主观权重值；
S06：计算各类上下文信息客观权重
所述主客观权重管理单元运用客观赋权算法，为需要分配权重的n类上下文信息分配权重，权重向量表示为
Figure PCTCN2017104988-appb-000002
Woi表示第i类上下文信息的客观权重值；
S07：判断是否有反馈产生
所述主客观权重管理单元判断是否有来自所述用户模块的用户反馈，若有，则执行步骤S08，若没有，则执行步骤S09；
S08：计算贡献率
所述主客观权重管理单元根据用户反馈的评估量化信息来计算主观权重和客观权重的贡献率，主观权重的贡献率为α，客观权重的贡献率为1-α，计算公式如式(Ⅰ)所示：
Figure PCTCN2017104988-appb-000003
式(Ⅰ)中，Ti为用户产生的反馈总数；Ri为：在有用户反馈的情况下，第i类上下文信息经融合推理后得到的结果与用户反馈一致的数目，此处一致是指用户认同系统通过融合推理后给出的结果，包括用户主动给出的认可评价(显式)或者根据用户表情及动作等间接 推断出的认可行为(隐式)；α的初始值设为0.5，即各类上下文信息的主观权重和客观权重的贡献率相同；
S09：主客观权重合成
所述主客观权重管理单元根据贡献率对各类上下文信息的主观权重和客观权重进行合成，得到每类上下文信息的主客观权重值，合成公式如式(Ⅱ)所示：
Figure PCTCN2017104988-appb-000004
式(Ⅱ)中，
Figure PCTCN2017104988-appb-000005
Wi为第i类上下文信息的主客观权重值；
最终得到的主客观权重为
Figure PCTCN2017104988-appb-000006
S10：多算法上下文信息融合
所述上下文信息融合单元结合主客观权重，用不同的上下文信息融合算法，分别对多类上下文信息进行融合，每种信息融合算法得出一个概率向量，用来表示系统所有感知结果的可能性，这几个相同或者不同的概率向量作为初级上下文信息用来进行上下文信息推理；
S11：上下文信息推理
所述上下文推理模块利用所述决策管理单元给出的推理算法，对步骤S10得到的初级上下文信息进行推理，几个相同或者不同的概率向量经过相应的推理算法推理后得到最终的感知结果，即可供所述上下文应用模块应用的高级上下文信息；
S12：上下文应用
所述上下文应用模块接收到高级上下文信息后，调整相应应用程序或设备，为用户提供恰当的服务；
S13：用户反馈
所述用户模块记录用户在使用上下文应用过程中对系统进行的调整或用户在使用系统过程中的行为，作为用户的显式反馈或者隐式反馈；
S14：评估量化
所述用户模块提取用户显式反馈或者隐式反馈，转化为对每类上下文信息准确性的判断信息，发送至所述主客观权重管理单元进行主观权重和客观权重各自贡献率的优化。
本发明的有益效果为：
1、不同类型的上下文信息可以反映同一事件的情况或状态，但是会对检测精度产生影响。并且，系统在为用户提供服务时，要充分考虑用户的主观偏好信息，做到“以人为本”。本发明所述基于反馈的自适应主客观权重上下文感知系统既考虑到用户的偏好信息，又考虑到了各类上下文信息对决策的客观影响，针对不同的感知事件选择合理的上下文信息种类，在上下文信息融合推理过程中为不同种类的上下文信息分配不同的权重因子，提高了上下文信息融合推理的效率和准确率，可以为用户提供更智能、个性化的服务；
2、本发明所述基于反馈的自适应主客观权重上下文感知系统结合用户的反馈量化信息动态更新优化各类上下文信息的主观权重和客观权重各自的贡献率，进而得到各类上下文信息的主客观权重，即主观权重和客观权重的优化组合值，提高了上下文感知系统的智能性和精确度，可以为用户提供更加个性化的服务。
附图说明
图1为本发明所述基于反馈的自适应主客观权重上下文感知系统的结构框架图；
图2为本发明所述基于反馈的自适应主客观权重上下文感知系统的工作流程图。
具体实施方式
下面结合说明书附图和实施例对本发明作进一步限定，但不限于此。
实施例1
一种基于反馈的自适应主客观权重上下文感知系统，如图1所示，包括原始上下文信息采集模块、上下文融合模块、上下文推理模块、上下文应用模块以及用户模块。
原始上下文信息采集模块与上下文融合模块相连，上下文融合模块、上下文推理模块、上下文应用模块以及用户模块依次循环连接；
原始上下文采集模块用于采取不同的方式从不同的信息源采集原始上下文信息，原始上下文信息是指从不同传感器采集到的原始数据；上下文融合模块用于对原始上下文信息进行融合处理，提取出可供所述上下文推理模块使用的初级上下文信息；上下文推理模块用于对各种不同的初级上下文信息进行推理，得出可供所述上下文应用模块直接使用的高级上下文信息；上下文应用模块利用所述高级上下文信息，调整相应应用程序或设备，为用户提供恰当的服务；用户模块提取用户显式或隐式反馈，经过量化评估后转化为对每类上下文信息准确性的判断信息，发送至所述主客观权重管理单元，用于对各类上下文信息进行主观权重和客观权重各自贡献率的优化；显式反馈是指用户主动做出的反馈行为，如采用满意度评分机制收集到的用户数据，隐式反馈是指上下文感知系统通过用户表情识别，用户行为分析间接 推断出来的用户数据。
上下文融合模块包括依次连接的上下文信息预处理单元、主客观权重管理单元、上下文信息融合单元；
上下文信息预处理单元对原始上下文信息进行数据建模，建模模式为“感知类型+感知信息”，并对原始上下文信息分类，通过均值方法、最小二乘方法、最大似然估计方法或卡尔曼滤波方法对同类上下文信息间进行缺失值处理和融合操作；
主客观权重管理单元用于：根据系统感知事件类型，确认需要分配权重的上下文信息种类，并用主观赋权算法计算各类上下文信息的主观权重，所述主观赋权算法包括：Delphi算法、循环打分法、二项系数法和层次分析法(Analytic Hierarchy Process，AHP)；然后用客观赋权法计算各类上下文信息的客观权重，所述客观赋权法包括主成分分析法(Principal Components Analysis，PCA)、变异系数法、熵权法和多目标规划法，根据用户反馈的评估量化信息动态计算各类上下文信息的主观权重和客观权重的贡献率，最后根据贡献率得出各类上下文信息的主客观权重；
上下文信息融合单元对所述主客观权重管理单元给出的多类上下文信息和主客观权重用多种信息融合算法进行融合，得到初级上下文信息，所述信息融合算法包括基于神经网络的融合算法，基于卡尔曼滤波的融合算法、基于模糊理论的融合算法和基于D-S证据理论的融合算法。
上下文推理模块包括依次连接的决策管理单元和上下文信息推理单元；
决策管理单元根据初级上下文信息的种类和感知事件类型选择合适的推理方法，并为多种信息融合算法的结果分配不同的权重以提高推理准确度；
上下文信息推理单元根据所述决策管理单元给出的推理方法，对所述上下文信息融合单元的不同融合算法产生的初级上下文信息进行推理处理，得到可供所述上下文应用模块直接使用的高级上下文信息，所述上下文推理方法包括本体推理方法、基于规则的推理方法、基于证据论的推理方法和基于贝叶斯网络的推理方法。
实施例2
实施例1所述的上下文感知系统的工作方法，如图2所示，本实施例以上下文感知计算的一个场景——智慧健身系统为例，在智慧健身系统中，通过不同传感器分别采集用户的体重、身高、心率和血氧饱和度这些上下文信息，分别表示为Iw、Ih、Ir、Io。通过采集这4种上下文信息，最终给用户提出合适运动强度的健身项目建议。此例中，最后给出的建议包 括三种：较弱强度的项目、中等强度的项目和较高强度的项目。包括步骤如下：
S01：原始上下文信息采集
获取的原始上下文信息有1组体重信息Iw，2组身高信息Ih1、Ih2，2组心率信息Ir1、Ir2，1组血氧饱和度信息Io。
S02：上下文信息建模
建模好的上下文信息为Iw＝“感知类型-体重”+“感知信息-70kg”，Ih1＝“感知类型-身高”+“感知信息-180cm”，Ih2＝“感知类型-身高”+“感知信息-180cm”，Ir1＝“感知类型-心率”+“感知信息-70次/min”，Ir2＝“感知类型-心率”+“感知信息-72次/min”，Io＝“感知类型-血氧饱和度”+“感知信息-95％”。
S03：上下文信息分类融合
对2组身高信息Ih1、Ih2进行处理后得到一组数据Ih，对2组心率信息Ir1、Ir2进行处理后得到一组数据Ir，此处无数据缺失且数据量较小，融合算法选择均值法。
S04：确认感知所需上下文信息类型
感知事件类型为用户提供合理运动强度的健身项目，确认忽略掉身高信息，需要分配的上下文信息种类为体重、心率、血氧饱和度，上下文信息种类总数为3。
S05：计算各类上下文信息主观权重
根据专家建议及用户喜好等信息，用主观赋权算法赋予三类上下文信息Iw、Ir、Io的权重向量为
Figure PCTCN2017104988-appb-000007
S06：计算各类上下文信息客观权重
用客观赋权算法赋予三类上下文信息Iw、Ir、Io的客观权重向量为
Figure PCTCN2017104988-appb-000008
S07：判断是否有反馈产生
判断是否有用户的反馈信息，即用户对于系统建议的运动方式是否适合自己的反馈信息。若接收到反馈信息，则执行步骤S08，若没有接收到反馈信息，则执行步骤S09；
S08：计算贡献率
n＝3，初始的主观权重贡献率为α＝0.5，当用户给出反馈后，根据公式(Ⅰ)计算主观权重的贡献率α和客观权重的贡献率1-α，公式为：
Figure PCTCN2017104988-appb-000009
S09：主客观权重合成
三类上下文信息Iw、Ir、Io的主观权重向量为
Figure PCTCN2017104988-appb-000010
客观权重向量为
Figure PCTCN2017104988-appb-000011
用公式(Ⅱ)对两种权重进行合成，公式为
Figure PCTCN2017104988-appb-000012
最终得到的主客观权重向量为
Figure PCTCN2017104988-appb-000013
S10：多算法上下文信息融合
结合最终给出的主客观权重向量，用4种不同的上下文融合算法(基于神经网络的融合算法，基于卡尔曼滤波的融合算法、基于模糊理论的融合算法和基于D-S证据理论的融合算法)对3种上下文信息Iw、Ir、Io进行融合，每个算法得出一个用概率向量表示的建议，如：第一种算法的融合结果为用户适合较弱强度的项目、中等强度的项目还是较高强度的项目各自的概率为(0.6,0.3,0.1)，共得出4个相同或者不同的概率向量。
S11：上下文信息推理
对步骤S10得出的4个概率向量用所述决策管理单元给出的推理算法进行推理最后给出最终的1个建议，如：该用户适合中等强度的健身项目；
S12：上下文应用
系统根据步骤S11给出的建议—用户适合中等强度的健身项目，自动在屏幕显示中等强度运动项目类型及相关注意事项。
S13：用户反馈
用户看到系统给出的运动建议，选择了中等强度运动的健身项目，则认为该次用户反馈信息与上下文感知系统的推理结果一致。
S14：评估量化
所述用户模块动态保存用户的反馈次数以及推理结果与用户反馈一致的数目，并将这些 信息实时发送至主客观权重管理单元动态更新主观权重和客观权重各自的贡献率。

Claims (4)

  1. 一种基于反馈的自适应主客观权重上下文感知系统，其特征在于，包括原始上下文信息采集模块、上下文融合模块、上下文推理模块、上下文应用模块以及用户模块；
    所述原始上下文信息采集模块与所述上下文融合模块相连，所述上下文融合模块、所述上下文推理模块、所述上下文应用模块以及所述用户模块依次循环连接；
    所述原始上下文采集模块用于采取不同的方式从不同的信息源采集原始上下文信息，所述原始上下文信息是指从不同传感器采集到的原始数据；所述上下文融合模块用于对原始上下文信息进行融合处理，提取出可供所述上下文推理模块使用的初级上下文信息；所述上下文推理模块用于对各种不同的初级上下文信息进行推理，得出可供所述上下文应用模块直接使用的高级上下文信息；所述上下文应用模块利用所述高级上下文信息，调整相应应用程序或设备，为用户提供恰当的服务；所述用户模块提取用户显式或隐式反馈，经过量化评估后转化为对每类上下文信息准确性的判断信息，发送至所述主客观权重管理单元，用于对各类上下文信息进行主观权重和客观权重各自贡献率的优化；所述显式反馈是指用户主动做出的反馈行为，所述隐式反馈是指上下文感知系统通过用户表情识别，用户行为分析间接推断出来的用户数据。
  2. 根据权利要求1所述的一种基于反馈的自适应主客观权重上下文感知系统，其特征在于，所述上下文融合模块包括依次连接的上下文信息预处理单元、主客观权重管理单元、上下文信息融合单元；
    所述上下文信息预处理单元对原始上下文信息进行数据建模，建模模式为“感知类型+感知信息”，并对原始上下文信息分类，通过均值方法、最小二乘方法、最大似然估计方法或卡尔曼滤波方法对同类上下文信息间进行缺失值处理和融合操作；
    所述主客观权重管理单元用于：根据系统感知事件类型，确认需要分配权重的上下文信息种类，并用主观赋权算法计算各类上下文信息的主观权重，所述主观赋权算法包括：Delphi算法、循环打分法、二项系数法和层次分析法；然后用客观赋权法计算各类上下文信息的客观权重，所述客观赋权法包括主成分分析法、变异系数法、熵权法和多目标规划法，根据用户反馈的评估量化信息动态计算各类上下文信息的主观权重和客观权重的贡献率，最后根据贡献率得出各类上下文信息的主客观权重；
    所述上下文信息融合单元对所述主客观权重管理单元给出的多类上下文信息和主客观权重用多种信息融合算法进行融合，得到初级上下文信息，所述信息融合算法包括基于神经网 络的融合算法，基于卡尔曼滤波的融合算法、基于模糊理论的融合算法和基于D-S证据理论的融合算法。
  3. 根据权利要求2所述的一种基于反馈的自适应主客观权重上下文感知系统，其特征在于，所述上下文推理模块包括依次连接的决策管理单元和上下文信息推理单元；
    所述决策管理单元根据初级上下文信息的种类和感知事件类型选择合适的推理方法，并为多种信息融合算法的结果分配不同的权重以提高推理准确度；
    所述上下文信息推理单元根据所述决策管理单元给出的推理方法，对所述上下文信息融合单元的不同融合算法产生的初级上下文信息进行推理处理，得到可供所述上下文应用模块直接使用的高级上下文信息，所述上下文推理方法包括本体推理方法、基于规则的推理方法、基于证据论的推理方法和基于贝叶斯网络的推理方法。
  4. 权利要求3所述的一种基于反馈的自适应主客观权重上下文感知系统的工作方法，其特征在于，包括步骤如下：
    S01：原始上下文信息采集
    从各类传感器获取原始上下文信息；
    S02：上下文信息建模
    所述上下文信息预处理单元对原始上下文信息进行数据建模，建模模式为“感知类型+感知信息”；
    S03：上下文信息分类融合
    所述上下文预处理单元对原始上下文信息进行分类，即表示同种属性的上下文信息归为同类原始上下文信息，同类原始上下文信息间，通过均值方法、最小二乘方法、最大似然估计方法或卡尔曼滤波方法进行缺失值处理和融合操作；
    S04：确认感知所需上下文信息类型
    所述主客观权重管理单元根据系统感知事件类型及各类上下文信息与该事件的相关性来确认需要分配权重的上下文信息种类，需要分配权重的上下文信息种类数量总数设为n；
    S05：计算各类上下文信息主观权重
    所述主客观权重管理单元运用主观赋权算法，根据用户偏好，为需要分配权重的n类上下文信息分配权重，权重向量表示为
    Figure PCTCN2017104988-appb-100001
    Wsi表示第i类上下文信息的主观权重值；
    S06：计算各类上下文信息客观权重
    所述主客观权重管理单元运用客观赋权算法，为需要分配权重的n类上下文信息分配权重，权重向量表示为
    Figure PCTCN2017104988-appb-100002
    Woi表示第i类上下文信息的客观权重值；
    S07：判断是否有反馈产生
    所述主客观权重管理单元判断是否有来自所述用户模块的用户反馈，若有，则执行步骤S08，若没有，则执行步骤S09；
    S08：计算贡献率
    所述主客观权重管理单元根据用户反馈的评估量化信息来计算主观权重和客观权重的贡献率，主观权重的贡献率为α，客观权重的贡献率为1-α，计算公式如式(Ⅰ)所示：
    Figure PCTCN2017104988-appb-100003
    式(Ⅰ)中，Ti为用户产生的反馈总数；Ri为：在有用户反馈的情况下，第i类上下文信息经融合推理后得到的结果与用户反馈一致的数目；
    S09：主客观权重合成
    所述主客观权重管理单元根据贡献率对各类上下文信息的主观权重和客观权重进行合成，得到每类上下文信息的主客观权重值，合成公式如式(Ⅱ)所示：
    Figure PCTCN2017104988-appb-100004
    式(Ⅱ)中，
    Figure PCTCN2017104988-appb-100005
    Wi为第i类上下文信息的主客观权重值；
    最终得到的主客观权重为
    Figure PCTCN2017104988-appb-100006
    S10：多算法上下文信息融合
    所述上下文信息融合单元结合主客观权重，用不同的上下文信息融合算法，分别对多类上下文信息进行融合，每种信息融合算法得出一个概率向量，用来表示系统所有感知结果的可能性，这几个相同或者不同的概率向量作为初级上下文信息用来进行上下文信息推理；
    S11：上下文信息推理
    所述上下文推理模块利用所述决策管理单元给出的推理算法，对步骤S10得到的初级上下文信息进行推理，几个相同或者不同的概率向量经过相应的推理算法推理后得到最终的感知结果，即可供所述上下文应用模块应用的高级上下文信息；
    S12：上下文应用
    所述上下文应用模块接收到高级上下文信息后，调整相应应用程序或设备，为用户提供恰当的服务；
    S13：用户反馈
    所述用户模块记录用户在使用上下文应用过程中对系统进行的调整或用户在使用系统过程中的行为，作为用户的显式反馈或者隐式反馈；
    S14：评估量化
    所述用户模块提取用户显式反馈或者隐式反馈，转化为对每类上下文信息准确性的判断信息，发送至所述主客观权重管理单元进行主观权重和客观权重各自贡献率的优化。
PCT/CN2017/104988 2016-12-30 2017-09-30 一种基于反馈的自适应主客观权重上下文感知系统及其工作方法 WO2018120963A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
JP2019532041A JP6771105B2 (ja) 2016-12-30 2017-09-30 フィードバックベースの自己適応主客観的重みコンテキストアウェアネスシステムおよびその動作方法。

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201611252003.XA CN106650937B (zh) 2016-12-30 2016-12-30 一种基于反馈的自适应主客观权重上下文感知系统及其工作方法
CN201611252003.X 2016-12-30

Publications (1)

Publication Number Publication Date
WO2018120963A1 true WO2018120963A1 (zh) 2018-07-05

Family

ID=58836300

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2017/104988 WO2018120963A1 (zh) 2016-12-30 2017-09-30 一种基于反馈的自适应主客观权重上下文感知系统及其工作方法

Country Status (3)

Country Link
JP (1) JP6771105B2 (zh)
CN (1) CN106650937B (zh)
WO (1) WO2018120963A1 (zh)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113257023A (zh) * 2021-04-13 2021-08-13 哈尔滨工业大学 一种l3级自动驾驶风险评估与接管预警方法及系统

Families Citing this family (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106650937B (zh) * 2016-12-30 2019-09-27 山东大学 一种基于反馈的自适应主客观权重上下文感知系统及其工作方法
CN108052973B (zh) * 2017-12-11 2020-05-05 中国人民解放军战略支援部队信息工程大学 基于多项眼动数据的地图符号用户兴趣分析方法
CN108875030B (zh) * 2018-06-25 2021-05-18 山东大学 一种基于层次化综合质量指标QoX的上下文不确定性消除系统及其工作方法
CN109711663B (zh) * 2018-11-15 2021-03-02 国网山东省电力公司淄博供电公司 基于大数据分析的变电站油浸式变压器状态评估与修正方法及系统
CN110599033A (zh) * 2019-09-12 2019-12-20 辽宁工程技术大学 一种引入更新因子的采空区自燃危险动态预测方法
CN110619466A (zh) * 2019-09-16 2019-12-27 卓尔智联(武汉)研究院有限公司 信息处理方法、装置及存储介质

Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20120059780A1 (en) * 2009-05-22 2012-03-08 Teknologian Tutkimuskeskus Vtt Context recognition in mobile devices
CN103246819A (zh) * 2013-05-20 2013-08-14 山东大学 一种面向普适计算的不一致性上下文消除系统和方法
CN104346451A (zh) * 2014-10-29 2015-02-11 山东大学 一种基于用户反馈的情景感知系统及其工作方法和应用
CN105095471A (zh) * 2015-08-07 2015-11-25 合肥工业大学 一种基于短记忆的上下文感知推荐方法
CN105389405A (zh) * 2015-12-30 2016-03-09 山东大学 一种多推理引擎融合上下文感知系统框架及其工作方法
CN105654134A (zh) * 2015-12-30 2016-06-08 山东大学 一种基于有监督自反馈的情景感知系统及其工作方法与应用
CN106650937A (zh) * 2016-12-30 2017-05-10 山东大学 一种基于反馈的自适应主客观权重上下文感知系统及其工作方法

Family Cites Families (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
SG135048A1 (en) * 2000-10-18 2007-09-28 Johnson & Johnson Consumer Intelligent performance-based product recommendation system
US20040025183A1 (en) * 2002-07-31 2004-02-05 Koninklijke Philips Electronics N.V. Optimization of personal television
JP5491891B2 (ja) * 2009-05-11 2014-05-14 パナソニック株式会社 機器マネージメント装置およびプログラム
US8886586B2 (en) * 2009-05-24 2014-11-11 Pi-Coral, Inc. Method for making optimal selections based on multiple objective and subjective criteria
JP5851737B2 (ja) * 2011-07-04 2016-02-03 大和ハウス工業株式会社 室内環境制御システム及び室内環境制御方法
JP5775417B2 (ja) * 2011-10-18 2015-09-09 Kddi株式会社 ユーザインタフェース自動分析評価システム及びユーザインタフェース自動分析評価方法
CN102594928B (zh) * 2012-04-05 2014-07-02 山东大学 一种协作上下文感知的框架模型

Patent Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20120059780A1 (en) * 2009-05-22 2012-03-08 Teknologian Tutkimuskeskus Vtt Context recognition in mobile devices
CN103246819A (zh) * 2013-05-20 2013-08-14 山东大学 一种面向普适计算的不一致性上下文消除系统和方法
CN104346451A (zh) * 2014-10-29 2015-02-11 山东大学 一种基于用户反馈的情景感知系统及其工作方法和应用
CN105095471A (zh) * 2015-08-07 2015-11-25 合肥工业大学 一种基于短记忆的上下文感知推荐方法
CN105389405A (zh) * 2015-12-30 2016-03-09 山东大学 一种多推理引擎融合上下文感知系统框架及其工作方法
CN105654134A (zh) * 2015-12-30 2016-06-08 山东大学 一种基于有监督自反馈的情景感知系统及其工作方法与应用
CN106650937A (zh) * 2016-12-30 2017-05-10 山东大学 一种基于反馈的自适应主客观权重上下文感知系统及其工作方法

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113257023A (zh) * 2021-04-13 2021-08-13 哈尔滨工业大学 一种l3级自动驾驶风险评估与接管预警方法及系统
CN113257023B (zh) * 2021-04-13 2022-09-09 哈尔滨工业大学 一种l3级自动驾驶风险评估与接管预警方法及系统

Also Published As

Publication number Publication date
JP6771105B2 (ja) 2020-10-21
CN106650937A (zh) 2017-05-10
CN106650937B (zh) 2019-09-27
JP2020504374A (ja) 2020-02-06

Similar Documents

Publication Publication Date Title
WO2018120963A1 (zh) 一种基于反馈的自适应主客观权重上下文感知系统及其工作方法
KR101254181B1 (ko) 하이브리드 방식의 영상 데이터 전처리 기법 및 방사형 기저함수 기반 신경회로망을 이용한 얼굴 인식 방법
Zhu et al. Multiple-facial action unit recognition by shared feature learning and semantic relation modeling
CN112116025A (zh) 用户分类模型的训练方法、装置、电子设备及存储介质
KR101676101B1 (ko) 동적보상퍼지신경네트워크(dcfnn)를 기반으로 한 얼굴인식 알고리즘
CN117598674B (zh) 多参数心脏功能监测系统及方法
Yeoh et al. Genetic algorithm assisted by a svm for feature selection in gait classification
Bhattacharjee et al. A comparative study of supervised learning techniques for human activity monitoring using smart sensors
Khan et al. Optimized features selection for gender classification using optimization algorithms
Huang et al. Calibration-aware bayesian learning
Wei et al. ActiveSelfHAR: Incorporating Self-Training Into Active Learning to Improve Cross-Subject Human Activity Recognition
CN115392302A (zh) 一种基于融合图卷积网络的脑电情绪识别方法
Kishore et al. A hybrid method for activity monitoring using principal component analysis and back-propagation neural network
Sravanthi et al. An efficient classifier using machine learning technique for individual action identification
CN112116024A (zh) 用户分类模型的方法、装置、电子设备和存储介质
Zhang et al. ECMER: Edge-Cloud Collaborative Personalized Multimodal Emotion Recognition Framework in the Internet of Vehicles
Fu et al. Image categorization using ESFS: a new embedded feature selection method based on SFS
WO2020191988A1 (zh) 新类别识别方法和基于模糊理论和深度学习的机器人系统
Lee et al. Bayesian personalized-wardrobe model (bp-wm) for long-term person re-identification
Chen et al. Federated multi-task hierarchical attention model for sensor analytics
Chen et al. Dual Attention Network for Unsupervised Domain Adaptive Person Re-identification
Sheng et al. CDFi: Cross-Domain Action Recognition using WiFi Signals
Yu et al. Deep meta-learning for personal thermal comfort modeling in office buildings
Babu et al. DCGAN-based Facial Expression Synthesis for Emotion Well-being Monitoring with Feature Extraction and Cluster Grouping
Bohra et al. Human Crime Based Intrusion Detection by Semantic Features Using LSTM with Inception Deep Learning Approach

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 17888550

Country of ref document: EP

Kind code of ref document: A1

ENP Entry into the national phase

Ref document number: 2019532041

Country of ref document: JP

Kind code of ref document: A

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 17888550

Country of ref document: EP

Kind code of ref document: A1