Public Facebook interaction analysis on timber as sustainable building material for climate change mitigation using AI

Ramit Debnath and Michael Ramage, Centre for Natural Materials Innovation, University of Cambridge   

Figure 1. Dataset from Facebook 

Analysed over 268,184 Facebook interactions from 1651 public posts on ‘timber’, ‘building materials’, ‘climate change’ between August 2020 and August 2021 (see Figure 1). The dataset covered a global scale that captured public reaction to trending posts and news on timber as a building material. Data was collected through Facebook’s CrowdTangle platform. Highest posts on these topics were observed from Australia, United Kingdom, United States and Zambia, while we other higher posts were aggregated as Rest of World (RoW) (see Figure 2).  

Figure 2. Post frequency and it distributions. 

Figure 3. Total number of interactions over time and the headlines of highest interacted posts 

The most interacted posts are presented below with word clusters for the countries with highest number of posts. It can be seen in Figure 3 that the intensity of the Facebook interactions gets dense from 06-2021, which can be a cumulative effect for COP-26, forest fires and built environment decarbonisation narratives. Further breakdown of these interactions and the associated posts is presented in Figure 4. It can be seen that for Australia the most widely used keywords were ‘build’, ‘timber’, ‘will use’, ‘hemp’, ‘construction’ and ‘land impact’. For United States, the most common keywords in the posts were ‘fire’, ‘forest’, ‘timber’, ‘build’, ‘materials’, ‘construction’ and ‘wood’. In the United Kingdom, the posts were concentrated around ‘timber’, ‘material’, ‘build’, ‘shipping’, and ‘sussex bridge’. Interestingly for India, a lot of emphasis on natural materials like ‘lime’, ‘bamboo’ and ‘timber’, with a range of distinctive words like ‘earthquake’, ‘absorb, ‘mortar’, ‘moisture’, ‘fabric’, etc (see Figure 4). The rest of the world denotes an aggregated highly repetitive keywords like ‘fire’, ‘wood’ ‘build’, ‘material’, ‘construction’ and ‘land’.   

Figure 4. Keywords from the posts in the respective countries 

The application of machine learning and AI was done through topic modelling using Latent Dirichlet Allocation (LDA) algorithm. LDA is an unsupervised machine learning-based classification technique that automatically analyses text data to determine cluster words from a set of documents. It is based on the basic idea that each document can be expressed as a distribution of topics, and each topic can be described by a distribution of words. We use the mathematical background for LDA as per the specifications of Blei. et al, 2003. The clustering results are presented below as per the countries mentioned above, where the ‘beta’ shows the probability of that particular words in the cluster and its relative importance. The algorithms are validated using state-of-the-art cross-validation techniques. 

The topic models for Australia shows that in cluster 1 (see Figure 5) high probability terms are associated with ‘timber’, ‘lightweight’, ‘build’, local’ and ‘construction’ which shapes the public narrative on timber being an important local lightweight building material. The topic 2 further includes ‘carbon’, ‘hemp’, low’ and ‘impact’ indicating that the Facebook posts were oriented towards the positive benefits of natural materials in construction for reducing carbon emissions. Topic 3 and topic 4 denote other elements of the Facebook interactions.  

Figure 5. Topic models for Australia 

Figure 6 illustrate the topic models for the United States. Topic cluster 1 and topic 3 indicates the contextualised interactions around forest fires in the country. Topic cluster 2 shows high probability words associated with ‘new material’, ‘construction’ and ‘build’. Topic cluster 4 presents irrelevant information that need further processing.  

Figure 6. Topic Models for United States 

Figure 7. Topic models for United Kingdom 

The topic models extracted for the United Kingdom is illustrated in Figure 7. Topic 1 and topic 2shows high probability words associated with construction, market and the built environment. It infers new market creation of low-carbon building materials, in which, timber is a critical element. Topic 3 and topic 4 further illustrates words like ‘recycle’, ‘material’, ‘ship’, ‘sea’, etc, indicating social media interactions around recyclability of timber and UK’s timber import/export industry.  

The topic models for India show discussions around natural materials like ‘lime’, ‘timber’ and ‘bamboo’ (see Figure 8, Topic 1). Topic 2 and topic 3 illustrates the importance of timber in post-disaster and post-covid construction industry. Whereas the topic 4 indicates the growing narratives on ‘learning from the west’ on sustainable use of engineered timber for decarbonising the Indian built environment, especially when it is estimated that 70% of the building stock is yet to be built.  

Figure 8. Topic models for India 

  • Jachthavenweg 111
    1081 KM Amsterdam
    The Netherlands
    +31 (0) 20 261 9688
    contact@builtbn.org

  • KVK: 83326901
    RSIN: 862830667
    Policy Plan
    Privacy policy

    © 2021 Built by Nature