Shuhan Fang, Ziyu Zhang, Yue Huang
Discover City Knowledge With Machine Studying | GSAPP
Professor: Mauricio Rada Orellana
Spring 2023
I. Introduction and Drawback Definition
Town of New York has lengthy been considered a world hub for tradition, vogue, and leisure. In recent times, it has additionally been acknowledged for its romantic attract, persistently rating excessive in varied romantic metropolis rankings. Moreover, the divorce charge in New York Metropolis has remained comparatively low in comparison with different main cities in the USA. These tendencies have piqued our curiosity and purpose to grasp why New York is taken into account a romantic metropolis and why its divorce charge is decrease than different cities.
To make clear these questions, our research will use a mixture of street-level picture evaluations, on-line media submit analyses, and sociodemographic traits to develop a Romantic Analysis Map. This analysis map will assist information the romantic design of cities by offering a deeper understanding of the components that contribute to romantic and secure city environments.
Utilizing deep neural networks, we are going to course of street-level photographs to acquire a semantically comprehensible description of the pictures, taking into consideration related objects and their placement within the picture. This evaluation will assist us perceive how individuals understand city areas and establish which options contribute to a romantic environment. We may also analyze the sociodemographic traits of on-line media submit writers and the content material of their posts to establish components that contribute to perceptions of romance in city environments.
Finally, our analysis goals to supply insights into the components that contribute to a metropolis’s romantic environment and the design ideas that may be utilized to create safer, extra romantic city environments. By growing a Romantic Analysis Method, we hope to information metropolis planners and designers in creating extra livable, romantic, and secure cities.
II. Literature Evaluate
The literature on the subject of romantic cities and concrete design is various and multi-disciplinary, incorporating fields akin to city planning, structure, psychology, and sociology. A number of research have examined the connection between the bodily parts of city landscapes and subjective perceptions of security and romance in city areas.
i) Romantic Index Definition
Perceptions of security and romance in city areas are advanced and influenced by a variety of bodily, social, and psychological components. Harvey (1979) notes that the complexity of city notion depends on two fundamental domains: the bodily parts of the panorama and their place in area, and the social and psychological subjective processes that interpret these parts. This means that the design of city areas just isn’t solely concerning the bodily parts but additionally concerning the notion and interpretation of these parts by people. Alvarez-Marin and Saldana Ochoa (2018) and Arefieva et al. (2018) each discover using geotagged knowledge to supply insights into the subjective experiences of people in city areas and the way this pertains to their preferences. Zhou et al. (2019) look at cultural variations in romantic conduct and supply a novel perspective on the subject of romance in city environments.
ii) Earlier Methodology
A number of research have examined using machine studying and massive knowledge evaluation to realize perception into perceptions of city area. Wu et al. (2020) discover using large knowledge and machine studying to measure the heterogeneous notion of city area, with a concentrate on security. The research highlights the potential of machine studying and massive knowledge to supply insights into the subjective experiences of people in city areas and the components that contribute to perceptions of security.
iii) Earlier Limitations
Earlier research lack consensus on what components contribute to a romantic environment and standardized strategies for measuring these components. Due to this fact, our proposed research goals to handle these limitations by utilizing a mixture of street-level picture evaluations, on-line media submit analyses, and sociodemographic traits to develop a Romantic Analysis Mannequin. This mannequin will present steerage for the romantic design of cities by figuring out the components that contribute to romantic and secure city environments. By growing a standardized methodology, our research will contribute to the literature by offering a extra complete and detailed understanding of the components that outline romantic city areas.
General, the literature means that perceptions of security and romance in city areas are advanced and influenced by a variety of bodily, social, and psychological components. Machine studying and geotagged knowledge evaluation are rising instruments that present new methods of understanding these perceptions and designing city areas which might be each secure and romantic.
III. Methodology
Within the current research, we mix the machine studying mannequin and statistical instruments to develop a methodological proposal that helps, within the lens of digital urbanism, to investigate and measure the romantic index of metropolis’s neighborhoods, and additional explores how social media influences metropolis engagement and the way this suggestions can inform and improve city planning methods.
i) Workflow
The proposed methodology, as proven in Determine.1, consists of 4 elements. First, knowledge assortment to extract romantic index key phrases and filter the picture dataset. We apply co-occurrence evaluation on the social media dataset to generate probably the most frequent key phrases of romantic metropolis. Then, picture parametrization utilizing PSPNet picture evaluation to phase and classify completely different parts on the street view photographs. Every picture within the dataset is labeled with “extra stunning/romance”, “livelier”, “safer”, “extra boring”, “extra miserable”, that are later quantified into values in knowledge preprocessing. Third, Fashions coaching with the dataset that concludes the romantic scores and parts proportion of labeled road view photographs. Mixed with parts proportion and label values, the fashions are educated to foretell the romantic index of different places. And lastly examine the efficiency of various fashions and visualize the outcomes with one of the best outcomes.
i) Datasets
1) Index Extraction
To extract key phrases associated to the romantic index of New York Metropolis, we collected tweet posts and Flickr feeds from 2021 to 2022. These knowledge sources present priceless insights into how individuals understand the town in a romantic approach, and permit us to establish the themes and matters which might be most intently related to romantic experiences in New York. In line with the phrase co-occurrence community from Tweet, as proven in Determine.2, we discover out the target romantic index with relative excessive frequency are: consuming, luck, place, purchasing, weekend, local weather. From the results of Flickr, as proven in Determine.3 , stunning, thoughts, journey, view, metropolis of goals are probably the most frequent key phrases. Concluded from the key phrases, it’s proven that typically, romantic index is expounded with structure, pure surroundings, eating and leisure, tradition and humanities, walkability, and so on.
2) Road view Picture
On this research, we leverage the Place Pulse 2.0 picture dataset as our main supply for picture evaluation attributable to its extensive number of city landscapes. We filter the dataset utilizing romantic key phrases and apply the PspNet picture evaluation strategy on the coaching dataset, as proven in Determine.4. The strategy to gather knowledge within the venture is thru using the PspNet picture evaluation. The PspNet algorithm makes use of a convolutional neural community to phase and classify the completely different parts on the street view picture, akin to buildings, sky, timber, roads, grass, and other people. The output of the PspNet algorithm is a CSV file that accommodates the proportion of every component current within the picture. The generated data helps us establish the important thing parts that contribute to the romantic index of a neighborhood. Mixed with labeled values of every picture, parts proportion generated from the PspNet picture evaluation strategy will probably be used because the coaching dataset for the following step of our analysis.
IV. Prediction Mannequin
The prediction mannequin that we have now constructed goals to foretell the romance rating of various neighborhoods based mostly on road view photographs and varied options. The mannequin constructing course of concerned a number of steps, together with knowledge preprocessing, function choice, mannequin coaching, and mannequin analysis. On this part, we are going to describe every of those steps intimately.
i) Construct the Mannequin
Step one in constructing our prediction mannequin was to preprocess the info. We transformed the qualitative labels for various options into numerical scores starting from 1 to five, after which used the ‘describe()’ technique to get a statistical abstract of the dataset’s options. (Fig.5 statistical abstract of the dataset) This helped us to grasp the distribution of the options and their. We then explored the correlation matrix (Fig. 6 correlation evaluation) to establish the relationships between completely different options and their potential impression on the goal variable. Based mostly on the correlation matrix, we chosen a subset of options for our mannequin (Fig 7 Function Chosen).
ii) Practice the Mannequin
Subsequent, we experimented with a number of machine studying fashions, together with KNN, SVM, Random Forest, Resolution Tree, OLS, and Gaussian. To judge the efficiency of those fashions, we used a number of metrics, together with R-squared, RMSE, and MAE. R-squared is used to measure the goodness of match of the mannequin, and it tells us how a lot of the variation within the goal variable might be defined by the unbiased variables. RMSE and MAE are used to measure the accuracy of the mannequin’s predictions, and so they inform us how far the anticipated values are from the precise values. (Fig. 8 Fashions Analysis Knowledge).
After evaluating the efficiency of various fashions, we chosen the Random Forest mannequin as one of the best mannequin for our prediction activity. (Fig. 9 Random Forest Regression Mannequin) We then educated the mannequin on the preprocessed dataset utilizing a coaching set. In the course of the coaching course of, we used a 10-fold cross-validation approach to keep away from overfitting and to make sure that the mannequin was generalizable to new knowledge. Moreover, we evaluated the significance of various options in predicting the romance rating. (Fig. 10 Function Significance).
iii) Prediction Mannequin
Utilizing the educated Random Forest mannequin, we are able to predict the romance rating of every neighborhood （Fig 11 romance rating predicted neighborhood) and streets (Fig 12 romance rating predicted streets ) based mostly on the options that we have now chosen. As soon as the predictions have been made, we saved the outcomes right into a CSV file for additional evaluation and visualization.
In conclusion, we have now constructed a strong prediction mannequin that may precisely predict the romance rating of various neighborhoods based mostly on road view photographs and varied options. The mannequin constructing course of concerned knowledge preprocessing, function choice, mannequin coaching, and mannequin analysis. We experimented with a number of machine studying fashions and evaluated their
efficiency utilizing a number of metrics. Our outcomes present that the Random Forest mannequin outperformed the opposite fashions by way of accuracy and generalizability. Our prediction mannequin might be helpful for city planners and metropolis officers to establish neighborhoods which might be extra romantic and engaging to guests and residents.
V. Outcomes
These prediction outcomes embrace the prediction map and the rating desk of all neighborhoods in Manhattan, NYC. It’s value noting that these rankings are based mostly on a specific set of things and will not essentially replicate the general desirability or livability of every neighborhood. Moreover, it’s essential to contemplate that perceptions of romance might range relying on particular person preferences.
General, the rating desk gives a helpful place to begin for additional exploration and evaluation of the components that contribute to perceptions of romance in city environments.
i) Prediction Map and Rating Desk
The highest-ranked neighborhoods, Higher East Aspect-Carnegie Hill and Turtle Bay-East Midtown, are each identified for his or her upscale residential areas and proximity to cultural establishments akin to museums and galleries, which may contribute to their larger romantic scores.
However, the lower-ranked neighborhoods, akin to Hudson Yards-Chelsea-Flatiron-Union Sq. and Gramercy, might have decrease romantic scores attributable to components akin to excessive ranges of economic growth, busy streets, and fewer inexperienced area.
ii) Neighborhoods
Take Midtown Manhattan, Chelsea and Central Park as examples. It is sensible that neighborhoods with extra facilities akin to shops, parks, and sights would have larger romantic scores, as these options can contribute to a extra nice and pleasant city surroundings.
Within the case of Midtown Manhattan, which incorporates standard vacationer sights akin to Occasions Sq. and the Empire State Constructing, the presence of those landmarks may contribute to a better romantic rating. Moreover, the neighborhood is residence to many high-end eating places, outlets, and inns, which may additionally contribute to a extra romantic environment.
Equally, Chelsea, which is understood for its artwork galleries, museums, and stylish eating places, may even have a better romantic rating as a result of presence of those cultural facilities. And as for Central Park, it’s no shock that this iconic park with its stunning landscapes and leisure actions would have a excessive romantic rating.
In all, the connection between the presence of facilities and romantic scores in these neighborhoods gives additional proof of the significance of city design and planning in creating engaging and livable city environments.
VI. Conclusion
On this research, the impossibility of really perceiving and evaluating a spot is addressed by predicting the extent of romance in Manhattan, New York. This impediment is dealt with by encompassing completely different communities, akin to satellite tv for pc photographs and road views, in an listed method concurrently. Every of those communities might be symbolized as a numerical vector, which is computed utilizing machine studying to construct and practice prediction fashions. We concentrate on characterizing locations from 4 important elements: location, public locations, transportation and occasions. As an indication, we make use of road view photographs and social media pictures as goal and subjective romantic index to implement the framework, and discover the romantic map of Manhattan.
i) Limitations
Limitations exist in three elements, first, knowledge extraction, second, predictive mannequin coaching, and third, analysis end result parsing. Firstly, Flichr image solely extracts some customers’ knowledge, ignoring the distinction of customers’ character, whereas the semantic key phrase extraction half is much less correct as a result of screening mechanism. Additional, the prognostic mannequin depends on PSPnet for scene key component disassembly, however the important thing parts and Romantic Index should not absolutely matched with one another. Within the ultimate scoring stage, the existence of neighborhood partitions for merging and disassembling in addition to the evaluation course of, the neglect of the density issue of Index results in scoring bias.
ii) Future Steps
Within the present part of this analysis, the ensuing utility ought to be thought of as a prototype, which is simply educated beneath machine studying based mostly on a portion of Flickr person photographs. Future steps on this analysis will probably be to open this pipeline to any person who desires to coach a private mannequin and acknowledge patterns of non-public preferences in city area.
VII. Bibliography
- Arefieva, V., Egger, R., & Yu, J. L. (2021). A machine studying strategy to cluster vacation spot picture on Instagram. Tourism Administration, 85, 104318. https://doi.org/10.1016/j.tourman.2021.104318
- Chen, C., Li, H., Luo, W., Xie, J., Yao, J., Wu, L., & Xia, Y. (2022). Predicting the impact of road surroundings on residents’ temper states in giant city areas utilizing machine studying and road view photographs. Science of the Whole Surroundings, 816, 151605. https://doi.org/10.1016/j.scitotenv.2021.151605
- Ramirez, T., Hurtubia, R., Lobel, H. O., & Galilea, P. (2021a). Measuring heterogeneous notion of city area with large knowledge and machine studying: An utility to security. Panorama and City Planning, 208, 104002. https://doi.org/10.1016/j.landurbplan.2020.104002
- Ramirez, T., Hurtubia, R., Lobel, H. O., & Galilea, P. (2021b). Measuring heterogeneous notion of city area with large knowledge and machine studying: An utility to security. Panorama and City Planning, 208, 104002. https://doi.org/10.1016/j.landurbplan.2020.104002
- Wang, J., Fan, Y., Palacios, J., Yuchen, C., Jeanrenaud, N. G., Obradovich, N., Zhou, C., & Zheng, S. (2022). World proof of expressed sentiment alterations through the COVID-19 pandemic. Nature Human Behaviour, 6(3), 349–358. https://doi.org/10.1038/s41562-022-01312-y
- Ye, Y., Zeng, W., Shen, Q., Zhang, X., & Lu, Y. (2019). The visible high quality of streets: A human-centred steady measurement based mostly on machine studying algorithms and road view photographs. Surroundings and Planning B: City Analytics and Metropolis Science, 46(8), 1439–1457. https://doi.org/10.1177/2399808319828734
- Zhang, F., Zu, J., Hu, M., Zhu, D., Kang, Y., Gao, S., Zhang, Y., & Huang, Z. (2020). Uncovering inconspicuous locations utilizing social media check-ins and road view photographs. Computer systems, Surroundings and City Programs, 81, 101478. https://doi.org/10.1016/j.compenvurbsys.2020.101478
- Zhang, S., Aktas, T., & Luo, J. (2021). Mi YouTube es Su YouTube? Analyzing the Cultures utilizing YouTube Thumbnails of In style Movies. In 2021 IEEE Worldwide Convention on Massive Knowledge (Massive Knowledge). https://doi.org/10.1109/bigdata52589.2021.9672037
