IoT and Engagement in the Ubiquitous Museum

Abstract

In increasingly hyper-connected societies, where individuals rely on short and fast online communications to consume information, museums face a significant survival challenge. Collaborations between scientists and museums suggest that the use of the technological framework known as Internet of Things (IoT) will be a key player in tackling this challenge. IoT can be used to gather and analyse visitor generated data, leading to data-driven insights that can fuel novel, adaptive and engaging museum experiences. We used an IoT implementation-a sensor network installed in the physical space of a museum-to look at how single visitors chose to enter and spend time in the different rooms of a curated exhibition. We collected a sparse, non-overlapping dataset of individual visits. Using various statistical analyses, we found that visitor attention span was very short. People visited five out of twenty rooms on average, and spent a median of two minutes in each room. However, the patterns of choice and time spent in rooms were not random. Indeed, they could be described in terms of a set of linearly separable visit patterns we obtained using principal component analysis. These results are encouraging for future interdisciplinary research that seeks to leverage IoT to get numerical proxies for people attention inside the museum, and use this information to fuel the next generation of possible museum interactions. Such interactions will based on rich, non-intrusive and diverse IoT driven conversation, dynamically tailored to visitors.

Keywords: IoT; mobile sensors; museum behaviour prediction; museum visitor analysis; space sensing; visitor attention; visitor engagement.

Figure 1

General overview of the research, from the collection of the data to their modelling, analysis and interpretation for planning purposes.

Figure 2

General overview of the museum. The view from the west side of the castle, image courtesy of Polo Museale delle Marche.

Figure 3

General arrangement of the museum: ground floor (a); and first floor (b) highlighting in blue the beacons’ locations and in red the position of two multimedia display.

Figure 4

Screen shots of the app running. From left to right: The home page, the list of room, a detail of the room, the user geo-localised in one of the rooms and the panoramic virtual tour.

Figure 5

Distribution of individual total visit length. External labels correspond to time intervals in minutes, while inner labels correspond to proportion of visits lasting a a time period within the matching interval. The vast majority of subjects $(80.4\%)$ completed the entire visit to the museum in less than half an hour. Significantly fewer visits $(13.7\%)$ had a total duration of 60 min or longer.

Figure 6

Cumulative proportion of number of rooms visited. The number of visited rooms varied from one to fourteen (out of twenty possible options). The average person visited between four and six rooms. In favour of high selectivity, notice that $40\%$ of the subjects visited up to three rooms, and that the majority of this segment actually visited only two rooms. However, $60\%$ of the subjects were moderate in selecting between four and fourteen rooms.

Figure 7

Number of unique visitors per room. In this figure we look at each room separately, showing the number of unique visitors each of them had. Four of the twenty rooms had $50\%$ or more of the visitors, while six rooms had less than $8\%$. These latter six rooms were removed from subsequent analyses. Concerning the ongoing story about selectivity, notice that there are four rooms that are favoured by visitors, and that there are ten rooms that received a sizeable proportion of visitors.

Figure 8

Length-of-visit distributions per room $\left(\mathrm{T}\right)$. Rooms could be roughly classified into two groups according to the time subjects spent in them. One group comprised rooms for which visitors spent one minute on average, with little variation, while the other group contained rooms for which the $\mathrm{T}$ median was between 1.5 and 2 min. They corresponded to distributions that also had larger dispersion (longer bars).

Figure 9

Cumulative proportion of room visits over time thresholds. A qualitative analysis of the curated content inside each room determined that a visitor would need at least three minutes for minimal content consumption. Our data show that approximately $75\%$ of room visits could only be classed as impressions.

Figure 10

Room visits considered impressions or consumptions using a three-minute threshold. Blue bars represent number of visits that lasted less than three minutes, while red bars represent the opposite. From previous analyses, we know that most visits stayed below four minutes, which means that the red bars represent both shallow consumptions and the $15\%$ of visits lasting beyond four minutes.

Figure 11

The PCA analysis of per-room length-of-visit data captured four meta-variables that described $75\%$ of the variance for the entire visitor dataset.