共查询到19条相似文献，搜索用时 488 毫秒
1.
信息技术的快速发展导致信息过载.推荐系统是解决信息过载最有效的方式之一.近年来,深度学习的快速发展也带动了推荐系统的进步,各种深度推荐算法层出不穷.然而由于候选物品数量巨大且用户兴趣动态变化,深度推荐算法的推荐复杂度巨大,难以在实际系统中单独使用.在深度推荐技术发展的同时,物品召回技术（也称近似搜索技术）也有了较大的发展与进步.本文先介绍基于距离最小化的物品召回的研究进展,再从向量索引、局部敏感哈希、哈希学习、向量量化四个方面来深入探讨基于内积最大化的物品召回技术的研究进展. 相似文献
2.
由于网络上每天有海量的新闻报道产生,新闻推荐已经成为减轻用户信息负载、实现个性化新闻信息获取的重要途径,并被广泛用于新闻网站和新闻APP中以提升用户体验.不同于传统的商品推荐,在新闻推荐中新的新闻文章产生速度很快,而且新闻的语义信息需要结合整体新闻文本去理解,给传统的基于ID和基于特征的推荐算法带来了很大的挑战.此外,用户的新闻阅读兴趣存在高度多样性和动态性的特点,使得准确的用户建模变得非常困难.本文介绍了一些基于深度学习的个性化新闻推荐算法,并探讨了新闻推荐未来的一些可行的方向. 相似文献
3.
尹琦 《南京气象学院学报》2010,(4):324-327
提出了一种基于模糊方法并考虑用户评价属性(相似度和可信度)的Web选择算法,从Web服务质量数据来源角度对服务质量的属性进行了分类和计算.对非功能质量属性,使用模糊控制的度量方法,将非功能属性值转换成0～1之间的实数值.使用反馈相似度和可信度相结合的方法,将最终的推荐权值推荐给用户.该方法的提出有助于提高Web服务选择结果的准确性. 相似文献
4.
推荐系统是电子商务领域最重要的技术之一,而协同过滤算法又是推荐系统用得最广泛的.提出了一种基于加权三部图网络的协同过滤算法,用户、产品及标签都被考虑到算法中,并且研究了标签结点的度对用户相似性计算的影响.实验结果表明,此算法在解决用户冷启动问题的同时,还具有较高的推荐准确性. 相似文献
5.
大众在旅游途中期望获得开销低、行程方便、舒适度高的旅游体验,同时还具有历史人文、自然景观、美食购物等不同游览需求.因此,本文提出了一种基于改进混合蛙跳算法的个性化旅游路线推荐方法.首先建立个性化旅游路线推荐问题的优化模型,并针对该模型的特点,设计改进混合蛙跳算法.通过调整可控精度,增加筛选准则和及时处理异常解等策略增强群体的多样性,降低遗漏最优解的风险,强化局部搜索能力,并提高算法的求解精度.以南京三日游个性化旅游路线推荐问题作为实例,收集南京市内知名景点的门票价格、开放时间、不同出行方式所需的时间和花费情况以及食宿费用等相关数据,基于改进混合蛙跳算法进行求解.实验结果表明,与改进前的方法相比,所提改进方法能够获取更优的路径解,推荐的路线能够更好地满足用户的个性需求. 相似文献
6.
传统的推荐系统主要针对单个用户,但随着社会和电子商务的快速发展,人们越来越多地以多个用户的形式一起参与活动,而群组推荐旨在为多个用户组成的群组提供服务,已成为当前研究的热点之一.针对目前群组推荐准确率低,群组成员之间偏好冲突难以融合的问题,本文提出了一种新的共识模型策略,融合了群组领袖影响因子和项目热度影响因子,基于K近邻为目标群组寻找邻居群组,借鉴邻居群组的偏好,设计了基于偏好融合的群组推荐算法.在MovieLens数据集上的实验结果表明,本文所提的融合策略较传统的偏好融合策略有着更优越的表现,推荐准确率（nDCG）的总体平均性能约提高13%,推荐列表多样性指标的总体平均性能约提高10%. 相似文献
7.
8.
使用传统协同过滤的方式进行推荐往往会忽视音乐底层特征.通过将音乐的音频特征与歌词信息进行多模态融合,并将融合后的特征信息作为协同过滤推荐的补充,提出了一种基于多模态的音乐推荐系统.主要探讨了音频特征与歌词信息的提取,并在提取歌词信息时利用LDA主题模型进行特征降维.针对多模态融合问题,使用一种特征级联早融合法（EFFC）融合方式,并将多模态融合后的结果与单模态结果进行了比较.对于结果的推荐,以多模态特征信息为依据建立用户兴趣模型,并将该模型通过LSTM神经网络,以过滤与优化协同推荐的用户组.结果表明,基于多模态的音乐推荐系统将推荐结果的误差项平方和（SSE）由传统的2.009降至0.388 6,验证了该方法的有效性. 相似文献
9.
【目的】面对日益增加的夏季电力气象服务用户需求和激烈的市场竞争,必须明确用户需求的优先顺序,为有序改善产品提供参考。【方法】利用Kano模型,通过问卷调查的方式收集了受试者对各项需求点的反馈,并计算B系数和W系数,绘制B-W象限图,得到17个需求点的需求类型和优先级排序。【结果】17个需求点可分为基本型（6个,35%）、期望型（4个,24%）、魅力型（4个,24%）和无差异型（3个,18%）4种类型。基本型最多,无差异型最少,各类型间界限较清晰。各需求点优先级顺序较明确,即用户首先要求产品应准时、内容完整和节点透明,其次要求产品应具备较高的准确率和个性化水平,最后要求产品的美观模块化程度便于高效地抽取使用并及时得到专家支持,而产品是否处于商用化阶段和是否有特色的优先级最差。但值得关注的是D16特色产品的需求点被视为魅力型需求的票数高达20票（共50票）,接近归类于魅力型需求（21票）。【结论】应以智能化业务系统建设和提升专家支持服务水平为工作核心来满足用户需求。 相似文献
10.
本文针对2009年6月5日发生在我国华东地区的一次中尺度对流过程（Mesoscale Convective System,简称MCS）,基于集合均方根滤波（Ensemble Square Root Filter,简称EnSRF）方法同化多部多普勒天气雷达资料,引入具有时空自适应理论优势的贝叶斯膨胀算法,通过与常数膨胀算法的对比,分析了两种膨胀算法对EnSRF同化效果的影响。结果表明:贝叶斯膨胀算法同化的雷达组合反射率因子在强对流中心处有所增强,改善了基于常数膨胀算法的EnSRF同化强对流系统偏弱的问题。相比常数膨胀算法,贝叶斯膨胀算法同化的冷池结构更合理,径向风和反射率因子的均方根误差均减少。进一步探讨贝叶斯膨胀算法对同化效果改善的原因,结果发现:贝叶斯膨胀参数的分布与反射率因子的均方根误差分布十分吻合,这表明贝叶斯膨胀算法可以在背景场均方根误差较大,即背景场与观测差距较大时,给出较大的膨胀参数,进而增加集合的背景场误差,使得观测权重增大,从而给出了较大的分析增量。对集合平均分析场进行了1小时的确定性预报发现,贝叶斯膨胀算法提高了预报模式对安徽与江苏交界处的强对流系统的模拟效果,回波强度更强,冷池强度和范围更大,且对于不同组合反射率因子的阀值,贝叶斯膨胀算法的评分（Equitable Threat Score,简称ETS）均高于常数膨胀算法。这表明贝叶斯膨胀算法有效地改进了基于常数膨胀算法的EnSRF同化雷达资料的效果。 相似文献
11.
Birgir Hrafnkelsson Jeffrey S. Morris Veerabhadran Baladandayuthapani 《Meteorology and Atmospheric Physics》2012,116(1-2):43-61
Bayesian hierarchical modeling is applied to the analysis of annual minimum and maximum temperatures. In both cases the generalized extreme value distribution is selected as the marginal distribution at each site due to its flexibility and theoretical basis. This distribution has three unknown parameters, a location parameter, a scale parameter and a shape parameter. The location and scale parameters are assumed to vary across sites while the shape parameter is assumed to be the same for all sites. The location and the scale parameters are modeled as two independent Gaussian spatial processes which are governed by the Matérn correlation function. These spatial processes play a central role in spatial predictions at unobserved sites. The location parameter is also modeled with a common time effect which includes a linear trend and a linear site effect with altitude, distance to open sea, latitude and longitude as covariates. This model allows for spatial predictions for any set of unobserved sites, the scale of the grid can be as fine as possible as long as the covariates are observed at each of the unobserved sites. Data on the annual minimum and maximum temperatures in Iceland from 1961 to 2009?at 72 sites are analyzed and used to predict the 2nd percentile of the minimum temperature for the 12-month period from July 2011 to June 2012 and the 98th percentile of maximum temperature for the year 2011 for a large set of unobserved sites across Iceland at which the covariates are available. The spatial predictions reveal that the 2nd percentile of minimum temperature for 2011 to 2012 reaches ?35°C in the central part of Iceland and ranges from ?12 to ?22°C around the coast of Iceland with the lowest coastal temperatures in the Northern part and the Northeastern part. The 98th percentile of maximum temperature in 2011 ranges from 10 to 30°C where the warmest areas are found in the Eastern part, the North-Eastern part and the Southern part of the country. The estimated increase in minimum and maximum temperatures over the years 1961 to 2009 is 0.71 and 0.47°C per decade, respectively, while the average annual temperature increased 0.24°C per decade over the same period. 相似文献
12.
A new framework for isolating individual feedback processes in coupled general circulation climate models. Part I: formulation 总被引：1，自引：1，他引：0
This paper proposes a coupled atmosphere–surface climate feedback–response analysis method (CFRAM) as a new framework for
estimating climate feedbacks in coupled general circulation models with a full set of physical parameterization packages.
The formulation of the CFRAM is based on the energy balance in an atmosphere–surface column. In the CFRAM, the isolation of
partial temperature changes due to an external forcing or an individual feedback is achieved by solving the linearized infrared
radiation transfer model subject to individual energy flux perturbations (external or due to feedbacks). The partial temperature
changes are addable and their sum is equal to the (total) temperature change (in the linear sense). The decomposition of feedbacks
is based on the thermodynamic and dynamical processes that directly affect individual energy flux terms. Therefore, not only
those feedbacks that directly affect the TOA radiative fluxes, such as water vapor, clouds, and ice-albedo feedbacks, but
also those feedbacks that do not directly affect the TOA radiation, such as evaporation, convections, and convergence of horizontal
sensible and latent heat fluxes, are explicitly included in the CFRAM. In the CFRAM, the feedback gain matrices measure the
strength of individual feedbacks. The feedback gain matrices can be estimated from the energy flux perturbations inferred
from individual parameterization packages and dynamical modules. The inter-model spread of a feedback gain matrix would help
us to detect the origins of the uncertainty of future climate projections in climate model simulations. 相似文献
13.
14.
R. A. Colman 《Climate Dynamics》2001,17(5-6):391-405
This study addresses the question: what vertical regions contribute the most to water vapor, surface temperature, lapse rate and cloud fraction feedback strengths in a general circulation model? Multi-level offline radiation perturbation calculations are used to diagnose the feedback contribution from each model level. As a first step, to locate regions of maximum radiative sensitivity to climate changes, the top of atmosphere radiative impact for each feedback is explored for each process by means of idealized parameter perturbations on top of a control (1?×?CO2) model climate. As a second step, the actual feedbacks themselves are calculated using the changes modelled from a 2?×?CO2 experiment. The impact of clouds on water vapor and lapse rate feedbacks is also isolated using `clear sky' calculations. Considering the idealized changes, it is found that the radiative sensitivity to water vapor changes is a maximum in the tropical lower troposphere. The sensitivity to temperature changes has both upper and lower tropospheric maxima. The sensitivity to idealized cloud changes is positive (warming) for upper level cloud increases but negative (cooling) for lower level increases, due to competing long and shortwave effects. Considering the actual feedbacks, it is found that water vapor feedback is a maximum in the tropical upper troposphere, due to the large relative increases in specific humidity which occur there. The actual lapse rate feedback changes sign with latitude and is a maximum (negative) again in the tropical upper troposphere. Cloud feedbacks reflect the general decrease in low- to mid-level low-latitude cloud, with an increase in the very highest cloud. This produces a net positive (negative) shortwave (longwave) cloud feedback. The role of clouds in the strength of the water vapor and lapse rate feedbacks is also discussed. 相似文献
15.
噪声是影响图像质量的重要因素,噪声的存在导致图像的某些特征细节不能被辨识以及图像的信噪比下降,所以图像去噪是图像处理中的一个重要问题.本文提出了基于小波框架的彩色图像的泊松去噪模型.在该模型中,赋权的l2项作为保真项,包含小波框架的l1项作为正则项.同时,又提出了解决该模型的重新赋权的split Bregman算法.最后,利用仿真实验以及PSNR（峰值信噪比）指标对该模型的去噪效果进行评估,评估结果表明该算法可行、有效. 相似文献
16.
基于区域活动轮廓模型在处理医学特定分割目标时,受到图像背景影响较大,难以提取目标边界轮廓.针对这一问题,提出一种改进LFI模型.该模型构造出局部适当图像来逼近原图像,且引入吸附因子约束曲线演化方向,恢复曲线演化渐进性,使得分割具有针对性.采用二值水平集方法实现整个分割过程,避免了传统水平集数值不稳定性.实验结果表明该方法可以快速有效地分割特定医学图像目标. 相似文献
17.
Simon Dietz 《Climatic change》2011,108(3):519-541
To what extent does economic analysis of climate change depend on low-probability, high-impact events? This question has received a great deal of attention lately, with the contention increasingly made that climate damage could be so large that societal willingness to pay to avoid extreme outcomes should overwhelm other seemingly important assumptions, notably on time preference. This paper provides an empirical examination of some key theoretical points, using a probabilistic integrated assessment model. New, fat-tailed distributions are inputted for key parameters representing climate sensitivity and economic costs. It is found that welfare estimates do strongly depend on tail risks, but for a set of plausible assumptions time preference can still matter. 相似文献
18.
Using the method of radiative ‘kernels’ an analysis is made of water vapour, lapse rate and ‘Planck’ (uniform vertical temperature) long wave feedbacks in models participating in the World Climate Research Program (WCRP) Coupled Model Intercomparison Project phase 3 (CMIP3). Feedbacks are calculated at climate change timescales from the A1B scenario, and at three ‘variability’ timescales from the corresponding preindustrial experiments: seasonal, interannual and decadal. Surface temperature responses show different meridional patterns for the different timescales, which are then manifest in the structures of the individual feedbacks. Despite these differences, mean water vapour feedback strength in models is positive for all models and timescales, and of comparable global magnitude across all timescales except for seasonal, where it is much weaker. Taking into consideration the strong positive lapse rate feedback at seasonal timescales, combined water vapour/lapse rate feedback is indeed similar across all timescales. To a good approximation, global water vapour feedback is found to be well represented by the temperature response along with an assumption of unchanged relative humidity under both variability and climate change. A comparison is also made of model feedbacks with reanalysis derived feedbacks for seasonal and interannual timescales. No strong relationships between individual modelled feedbacks at different timescales are evident: i.e., strong feedbacks in models at variability timescales do not in general predict strong climate change feedback, with the possible exception of seasonal timescales. There are caveats on this (and other) findings however, from uncertainties associated with the kernel technique and from, at times, very large uncertainties in estimating variability related feedbacks from temperature regressions. 相似文献
19.
We estimate climate sensitivity from observations, using the deseasonalized fluctuations in sea surface temperatures (SSTs) and the concurrent fluctuations in the top-of-atmosphere (TOA) outgoing radiation from the ERBE (1985–1999) and CERES (2000–2008) satellite instruments. Distinct periods of warming and cooling in the SSTs were used to evaluate feedbacks. An earlier study (Lindzen and Choi, 2009) was subject to significant criticisms. The present paper is an expansion of the earlier paper where the various criticisms are taken into account. The present analysis accounts for the 72 day precession period for the ERBE satellite in a more appropriate manner than in the earlier paper. We develop a method to distinguish noise in the outgoing radiation as well as radiation changes that are forcing SST changes from those radiation changes that constitute feedbacks to changes in SST. We demonstrate that our new method does moderately well in distinguishing positive from negative feedbacks and in quantifying negative feedbacks. In contrast, we show that simple regression methods used by several existing papers generally exaggerate positive feedbacks and even show positive feedbacks when actual feedbacks are negative. We argue that feedbacks are largely concentrated in the tropics, and the tropical feedbacks can be adjusted to account for their impact on the globe as a whole. Indeed, we show that including all CERES data (not just from the tropics) leads to results similar to what are obtained for the tropics alone — though with more noise. We again find that the outgoing radiation resulting from SST fluctuations exceeds the zerofeedback response thus implying negative feedback. In contrast to this, the calculated TOA outgoing radiation fluxes from 11 atmospheric models forced by the observed SST are less than the zerofeedback response, consistent with the positive feedbacks that characterize these models. The results imply that the models are exaggerating climate sensitivity. 相似文献