Virtual community

Communication related to mental health in a virtual community: text mining analysis of a digital exchange platform during the Covid-19 pandemic | BMC Psychiatry

Once the platform was made public, the number of registered users as well as unregistered daily visitors increased continuously. As of May 4, 2020, nearly 700 registered users are registered. In early April, there were up to 10,000 page views per day. However, the number of website visits fell sharply in April. During the last week of the evaluation (April 28 to May 4, 2020), an average of 795 views per day were recorded (min. 62; max. 1159). An evaluation by username and activity also revealed that for this portion of the 2,300 visits, the majority were passive consumers of the content and did not participate in the dialogue. As registration only required a valid email address and name, no further claims can be made about the users. The analysis was based on an adjusted dataset of 31,764 words.

Frequencies, n-grams and correlations

Figure 1 shows the most frequently used words (not > 75). The most used reference word was ‘good’ (not = 396), although in preparation, greetings (e.g. hello) and signature words (e.g. goodbye) were removed. Subject-specific words such as ‘crown’ (not = 144) and ‘virus’ (not= 100) are apparent. The words ‘crisis’ (not = 82) and ‘anxiety’ (not= 98) give a first indication of possible discussion points in forums and live chats.

Fig. 1

Wordcloud containing the most frequently used words (not > 75)

Analysis of bigrams (Fig. 2), filtered according to a frequency of not > 5, showed several clusters, the majority of which consisted of two constructs combined, such as the topics prevailing at that time in the media: for example, ‘Keep your distance’ Where ‘home office’. building ‘crown’was a nodal point that was associated, on the one hand, with stressful subjects and, on the other hand, with a cluster consisting of supporting (i.e. empathetic) content. Participants discussed the topic of ‘solitude’ and the impact of ‘crisis’ on ‘Mental Health‘. ‘Time’ also forms a nodal point between stressful topics and supportive topics. Thereby, ‘time’ was experienced during the test phase as a ‘hard situation’that must be ‘taken seriously’ . Participants also shared advice on coping strategies, how ‘take time’ at ‘do yourself good’ Where ‘go outside’and ‘go for a walk’ .

Figure 2
Figure 2

Bigram visualization in forums and live chats (not> 5)

In addition to the visualization of the bigrams, the correlation analysis (phi coefficient) shows that certain subjects often come together but do not follow each other directly in order, as is the prerequisite for the analysis of the n-grams ( table 1). Here too, the challenges faced by participants during the first wave of Covid-19 are revealed as ‘anxiety’ Due to ‘pandemic’ or the ‘virus’ .

Table 1 Second word correlations

On the other hand, there are also possible coping strategies, such as ‘relaxation’ and ‘sports’Where ‘listen’and ‘hear a voice’ which in this context is not discussed as a symptom, but as a coping strategy.

Analysis according to bigrams and trigrams, with an emphasis on advice on coping strategies with words ‘idea’ Where ‘point’ not limited to frequency, revealed a variety of activities, such as sports (eg walking, jogging), games (eg board games, Lego), writing (eg , letters, diary), reading and talking with relatives and affected people (e.g. forum, phone calls, WhatsApp) who have recommended each other.

Sentiment analysis

Most of the words did not describe neutral feelings. The results show a low tendency for positive sentiment between users, with a polarity varying between negative (−1) and positive (+1) with an average value of 0.04, in which words with a positive classification were represented with a preponderance of 72% (3737 against 1425). Table 2 summarizes the 15 most commonly used positive and negative words. Word ‘good’ (not= 396) with positive bias was the most frequently used, followed by ‘to help’(not= 156). The most frequently used word with a negative bias was ‘anxiety’(not= 98), followed by ‘crisis’(not= 82).

Table 2 Sentiment analysis with the 15 most common sentiments

In-depth sentiment analysis focused on topical words such as ‘solitude’, ‘crisis’, ‘crown’and ‘time’could show that every word, whether it seems to have a negative or positive connotation, carries both polarities (see Fig. 3). For example, the word ‘solitude’is related to ‘insulation’and also to ‘to be creative’. Since the polarity of feelings can range from −1 to 1, it appears that relatively strong negatively biased words were used in comparison to positively biased words.

Figure 3
picture 3

Sentiment analysis with previous words