摘要
由于全球化趋势,房产投资不再局限于某个地区,越来越多的人开始进行海外房产投资,但由于出行问题以及对当地市场的不熟悉,造成了对房产价值的认知偏差。为了克服以上缺点,提出了一个多实例深度排序与回归(MDRR)网络,用于对房屋进行可视化评估,其目的是从房屋的照片和文字描述(如卧室数量)来预测房屋的价值。该网络使用弱监督数据进行训练,不需要密集的人工标注。同时还设计了一组人类启发式方法,通过对解决方案空间施加限制来提升深层特征,通过实验表明,提出的方法能够较为准确的评估房产价值。
Due to the tends of globalization,real estate investment is no longer limited to a certain area,more and more people are starting to invest in overseas real estate,but due to travel problems and unfamiliarity with the local market,it has caused a cognitive bias in the value of real estate.In order to overcome the above shortcomings,a multi-instance deep sorting and regression(MDRR)network is proposed here for visual evaluation of the house.Its purpose is to predict the value of the house from the house’s photos and text description(such as the number of bedrooms).The network uses weakly supervised data for training and does not require intensive manual annotation.At the same time,a set of human heuristic methods are also designed to enhance the deep features by imposing restrictions on the solution space.Experiments show that the proposed method can more accurately evaluate the value of real estate.
作者
谢志伟
XIE Zhiwei(Department of Computer Engineering,Dongguan Polytechnic,Dongguan 523808,China)
出处
《微型电脑应用》
2021年第1期36-39,43,共5页
Microcomputer Applications
基金
广东省科技局计划项目(2016B050502001)
东莞职业技术学院政校行企合作开展科研与服务项目(201817)。
关键词
深度学习
回归
房产评估
多实例学习
deep learning
regression
property assessment
multi-instance learning