1# RECORD 2580ac3af513b0ee5f2e19ed10bd991f2a8c80152# TITLE: 3# YEAR: VENUE: DOI: 4# AUTHORS: 5# PAGES: 23 CANDIDATE TABLE PAGES (1-based): [16]6# ABSTRACT: 789===== PAGE 1 (score 8, numeric-paren 0, table-words 0, var-words 8) =====10111 12 13Extracting real estate values of rental apartment floor plans 14using graph convolutional networks 15 16Atsushi Takizawa 17Osaka Metropolitan University 18 19Keywords 20Rental apartment, rent, hedonic model, floor plan, access graph, graph convolutional networks 21 22Abstract 23Access graphs that indicate adjacency relationships from the perspective of flow lines of rooms are extracted 24automatically from a large number of floor plan images of a family-oriented rental apartment complex in Osaka 25Prefecture, Japan, based on a recently proposed access graph extraction method with slight modifications. We define 26and implement a graph convolutional network (GCN) for access graphs and propose a model to estimate the real 27estate value of access graphs as the floor plan value. The model, which includes the floor plan value and hedonic 28method using other general explanatory variables, is used to estimate rents, and their estimation accuracies are 29compared. In addition, the features of the floor plan that explain the rent are analyzed from the learned convolution 30network. Therefore, a new model for comprehensively estimating the value of real estate floor plans is proposed and 31validated. The results show that the proposed method significantly improves the accuracy of rent estimation compared 32to that of conventional models, and it is possible to understand the specific spatial configuration rules that influence 33the value of a floor plan by analyzing the learned GCN. 34 35Introduction 36Attempts have been made to evaluate housing preferences and satisfaction statistically and quantitatively, although 37they are subjective. Indices such as neighborhood, location, size, and housing type have been used frequently (Adir 38et al. 1996, Jim and Chen 2009, Lovejoy et al. 2010, Michelson 1977, Valente et al. 2005). Some studies evaluated 39houses based on more detailed attributes such as the number of rooms, size, shape, room size, and facilities (Hofman 40et al. 2006, Goodman et al. 2007). The latter study hypothesized that housing preferences are influenced by a 41combination of many detailed factors such as design, usability, and spatial configuration, and this may be true despite 42individual differences. However, quantifying these factors is difficult when they are considered in detail. For example, 43Jirovec et al. (1984) applied the building evaluation index proposed by Marans and Spreckelmeyer (1982) to evaluate 44housing for the elderly and attempted to explain it using a hedonic model (Rosen 1974) that includes aesthetic 45attributes. However, they pointed out that the explanatory power of the location variables was strong and that of the 46aesthetic ones was weak. Hofman et al. (2006) and Goodman et al. (2007) evaluated the number of rooms, area, and 47defined shape, which are relatively easy to quantify, as independent variables. It is difficult to quantify space at a 484950===== PAGE 16 (score 12, numeric-paren 6, table-words 0, var-words 6) =====515216 53 54Analysis 3: Understanding the overall estimation basis of the FPV model 55We characterized the estimation basis of the overall FPV model for all unique access graphs. IG is a relative value 56calculated for each floorplan, and it is not sensible to directly compare or aggregate the IG values of components 57with those of other floorplans. Therefore, the IG values of each floor plan component were standardized to have a 58mean of 0 and standard deviation of 1, and their distribution was tabulated for each component of all floor plans. The 59distributions were analyzed for significant differences from the overall mean at a significance level of 0.05 using the 60mean analysis method (ANOM; Nelson et al. 2005). The results are shown in Figure 15. Each point represents the 61mean value, and if the value is outside the range of the confidence interval [LDL, UDL], it is significantly different 62from the overall mean. The confidence interval decreases with an increase in the number of components. 63The node with the largest positive contribution to the FPV is the balcony (0.463), whereas nodes with the largest 64negative contribution are the Japanese room (−0.267), dining kitchen (−0.216), and entrance (−0.165). The presence 65of a balcony may be interpreted as a guarantee of the minimum FPV because a balcony is an essential place to dry 66laundry in a family-oriented floor plan. As the popularity of Japanese rooms is declining in Japan, it is understandable 67that its presence has a negative impact. The dining room, kitchen, and entrance are common to all properties and are 68not involved in the differentiation factor if they are only present; therefore, it is difficult to interpret them as anything 69other than an overall adjustment factor. 70For edges, the expression “bl-bt” for the edge type indicates that the edge connects the balcony and the bathroom. 71The edges with absolute values greater than 0.1 beyond the confidence interval are in the descending order from the 72highest positive contribution, closet-hallway (0.175), closet-dining kitchen (0.167), hallway-toilet (0.164), dining 73kitchen-Japanese room (0.152), and closet-entrance (0.121). The contribution tends to be higher when the closet is 74located at the hallway or entrance. The contribution is higher when the toilet is independent and entered from the 75hallway rather than that from the washroom. The Japanese room is entered from the dining room and kitchen rather 76than from the main hallway, which is the characteristic of modern layouts. Although these connections are convincing, 77the overall results suggest that the relationship between detailed accessory functions such as closets and toilets, 78instead of the relationship between rooms, tends to affect rent. The only significant edge with a negative contribution 79of −0.1 or less was the Japanese room-Japanese room (−0.138), which was less significant than the positive case. 80This result is easy to understand, although it suggests that the contribution of Japanese rooms tends to be lower when 81they are connected than when they stand alone, because of synergistic effects. 82 838485===== PAGE 17 (score 10, numeric-paren 0, table-words 0, var-words 10) =====868717 88 89 90Figure 15. Distribution of node and edge contributions for all unique access graphs and confidence intervals for the 91mean analysis method. 92 93Discussion 94Based on the proposed method and its results, we discuss the accuracy of the access graph data, relevance of the FPV 95model to human evaluations, and difference in use between the proposed method and an approach that makes the 96graph features explicit. 97We discuss the accuracy of the access graph data. In this study, we improved the accuracy of access graph 98recognition by making detailed improvements to the methods used in previous studies; however, there are still cases 99of misclassification. In this method, the accuracy of semantic segmentation significantly affects the accuracy of 100neighboring graph extraction. Segmentation fails in small areas such as bathrooms and closets, when the text in the 101description is prominent and when the floor plan is black or white. The author personally heard that the accuracy of 102the semantic segmentation was improved in the newer version of the graph extraction method compared to the one 103we used; however, even so, perfect extraction may still be difficult to achieve. There are rare cases where the floor 104plan of a property is not written correctly in the first place. There are rare images of floor plans in which there are no 105doors or windows and the space is divided. Extracting access graphs from these images results in disconnected images, 106and continuous improvement in the quality of the access graph is required. 107We explain the relationship between the FPV model and human evaluation. For example, the FPV and human 108evaluation may be similar in some respects and different in others, as seen in the aforementioned lack of detection of 109water circulation. The current GCN model does not appear to clearly detect the features of long flow lines beyond 110room pairs. However, this may be attributed to the visualization quirks of IG (Shrikumar et al. 2017), which is a 111general-purpose method such as IG but can output more intuitive results. GNNExplainer (Ying et al. 2019) is an 112explanatory method dedicated to GCNs. In this study, we used IG, which is theoretically superior and easy to 113implement; however, we believe that appropriate methods for explaining the GCN of floor plans need to be compared 114and investigated. The current access graphs represent only the connections between rooms with different uses. 115However, the presence or absence of openings was not considered except for the approach to the balcony. More basic 116features such as the size and shape of the rooms were not considered; it is highly possible that these points are the 117