Text mining for occupant perspectives on the physical workplace

Abstract

An analysis is presented of occupants' responses from a commercial building indoor environment satisfaction survey database. Building from satisfaction ratings and standardized categorical responses collected in surveys for 192 office buildings in the United States, text analysis software is used to analyse text responses to open-ended survey questions, focusing on occupants' perspectives on the workplace and building overall, temperature, and acoustics. These occupant texts detail interactions between occupants and their physical environment in a technical sense, but also interpret these interactions, assess their consequences, and reflect on social relationships and other matters that lie outside dimension-by-dimension assessments of the physical environment. Viewed together these texts reveal a user-centred perspective that points to issues that rest below the surface of more technical analyses of buildings, such as over air-conditioning, worker stress and frustration, workplace usability, and relationships between physical and other aspects of the workplace. Attending to this perspective could lead to improvements in occupant experience, building and technology design, building operations, and survey research, as well as inform initiatives that require occupant adaptation and cooperation, including those for reductions in greenhouse gas emissions.

Il est présenté une analyse des réponses des occupants provenant d'une base de données d'enquêtes de satisfaction sur l'environnement intérieur d'immeubles commerciaux. En s'appuyant sur les notations de satisfaction et les réponses normalisées par catégorie recueillies dans des enquêtes menées sur 192 immeubles de bureaux aux Etats-Unis, un logiciel d'analyse de texte est utilisé pour analyser les réponses textuelles aux questions ouvertes d'enquête, en se concentrant sur les points de vue des occupants sur le lieu de travail et l'immeuble dans son ensemble, la température et l'acoustique. Les textes de ces occupants détaillent les interactions entre les occupants et leur environnement physique dans une acception technique, mais interprètent également ces interactions, en évaluent les conséquences et réfléchissent aux relations sociales et aux autres questions qui se situent hors des évaluations dimension par dimension de l'environnement physique. Considérés dans leur ensemble, ces textes sont révélateurs d'un point de vue centré sur l'utilisateur qui met l'accent sur des problèmes qui ne parviennent pas à affleurer dans les analyses plus techniques des immeubles, tels que l'excès de climatisation, le stress et les frustrations des travailleurs, l'utilisabilité du lieu de travail, et les relations entre les aspects physiques et les autres aspects du lieu de travail. Le fait de porter attention à ce point de vue pourrait conduire à des améliorations en termes d'expérience vécue par les occupants, de conception des immeubles et des technologies, d'exploitation des immeubles, et d'enquêtes par sondage, et pourrait également influer sur des initiatives qui exigent l'adaptation et la coopération des occupants, notamment des initiatives favorisant la réduction des émissions de gaz à effet de serre.

Mots clés: performances des immeubles qualité environnementale intérieure adaptation des occupants perceptions des occupants analyse de texte satisfaction sur le lieu de travail

Keywords:

• building performance
• indoor environmental quality
• occupant adaptation
• occupant perceptions
• text analysis
• workplace satisfaction

Acknowledgements

The authors would like to thank Fergus Nicol, Nan Wishner and the anonymous Building Research & Information reviewers of this paper.

Notes

CBE traditionally uses a minimum response rate of 35% for including surveys in database benchmarking comparisons to prevent the inclusion of samples that may be very unrepresentative. The average response rate among these buildings is 49%. As in any survey where the sample is inherently non-random, representation of the population is not guaranteed and cannot be proven without probing non-respondents. A study reviewing a variety of customer satisfaction surveys, however, found no statistically significant correlation between response rate and reported satisfaction (Peterson and Wilson, 1992), nor is there such a correlation in the CBE surveys.

There are also free tools available for text analysis such as Rapid Miner and Electronic Lexical Knowledge Base (ELKB).

Most opportunities to comment in the surveys analysed arise after an environmental factor has been rated negatively, so responses are already structured as to basic sentiment. This study did not fully assess how ‘smart’ available natural language-processing techniques can be for analysing occupant texts. At the least, text data management capabilities – identifying and extracting keywords, recognizing misspellings, applying custom dictionaries, the ability to display all texts with certain characteristics, and presenting frequencies and co-frequencies (counts of texts that have two or more co-occurring keywords or themes) – were valuable, even while interpretation was based on reading.

Some of the uncategorized responses could undoubtedly fit in one or more of these 50 themes, while others might suggest new themes. Unclassified responses include no comment statements like ‘none’, clarifying statements such as ‘I just moved into this office’ or ‘satisfied with this building compared to typical possibilities, not in and of itself’. Most were just difficult to classify automatically, such as ‘love the location!’ or ‘the geese are a problem’, or where generic words (‘workspace’ or ‘office’) were the only readily extractable terms. Though with more work these statements could be assigned to a theme, doing so seemed unlikely to add much insight and it was unnecessary given the non-statistical tactic taken.

Examples of the other 50 themes include way finding, accessibility for disabled people, energy use, storage space, weather, childcare, gyms, aesthetics and art, storage, public transportation, computers, density, and productivity. As is (and perhaps inevitably), there is a bias toward identifying issues that are easy to see in keywords, e.g. the words associated with restrooms discussed below.

The question asked is ‘How would you best describe the sources of this discomfort?’, followed by a list of options. The percentages pertain only to occupants who provided a response to the checkbox question. CBE survey instruments often go into additional detail about temperature and thermal environment, e.g. asking whether the problem is ‘too hot’ or ‘too cold’ by season.