Activating Social Listening to Protect Online Reputation (part 2)

EDITOR’S NOTE: this is a guest post by Leopoldo Martínez D., a researcher, consultant on social media corporate intelligence and lecturer at UCV and IESA (Venezuela), and it was originally published on his blog (in Spanish).

 

  1. Introduction

As I set out in the first part of this post, I will show the evaluation of the online reputation in a real situation related to the tourism industry.

  1. The unexpected event: Shooting at music festival in Playa del Carmen, Riviera Maya

Between January 6 and 15, 2017, a music festival was scheduled to be held in Playa del Carmen, as well as a series of events related both to music and to the tourism industry. On January 15, a shooting occurred in a well-known bar where people were celebrating the closure of the festival.

When the shooting happened, alarm messages began to be spread through social networks to give information on the incident, as well as comments on the context in which the incident had occurred. Some conversations revealed an interesting fact: the shooting was not an isolated event, but it had its origin in the “situation of crime that the Riviera Maya was witnessing in 2011″.

Could this affect the Riviera Maya’s reputation as a tourist resort? Could it be that “several years in a situation of crime” had already influenced the tourism industry’s image? These are some of the questions that the public and private actors that provide services and products in this tourist area might be asking themselves.

  1. Evaluation of tourism industry’s online reputation in the Riviera Maya

Conversations were extracted from Twitter, considering the following scenarios:

  • Scenario 1 (January 15 shooting): conversations that occurred between January 16 and 31, 2017 about the shooting at the music festival.
  • Scenario 2 (Crime in the tourist area): conversations that occurred between December 01, 2016 and January 31, 2017 about the issue of criminality in the Riviera Maya.

Below you can see the most relevant aspects of each phase of the evaluation of the reputation:

  • Monitoring and Filtering: an ad hoc tool was used to extract data from Twitter’s servers. The first filtering was done with Excel, eliminating duplicate conversations and retweets.
  • Analysis and filtering: we used a tool by company MeaningCloud. It consists of an add-in integrated into Microsoft Excel that offers capabilities to perform sentiment analysis and text classification, among others.

First, sentiment analysis was carried out. This analysis outputs five polarities: Positive plus (P+), Positive (P), Neutral (NEU), Negative (N) and Negative plus (N+). Five polarities instead of three, as discussed in the first part of this post, give more detail about the sentiment that underlies a message. Those messages that did not have a polarity assigned were removed. The results of all messages obtained with their respective polarities are the following:

These graphs show that in both scenarios there are more conversations with negative polarity (E1: (77.41+15.41)% = 92.52%, E2: (17.97+50.38)% = 68.36%), which means that both the shooting at the music festival and criminality in the Riviera Maya are perceived with great concern.

Then, text classification was performed. For this purpose, MeaningCloud’s Social Media model was used first to classify texts coming from social media. The categories identified enabled to have a first insight of the types of conversations that emerged in both scenarios. However, for specific conversations as the ones of the scenarios 1 and 2, MeaningCloud gives the possibility of building in an easy way (without programming) a custom model with user-defined categories, taking as classification criteria the texts, phrases, and words that appear in the extracted data.

For the case of the assessment of the online reputation, this functionality is appropriate to define categories that classify texts that deal with the causes or effects of the treated problem. In this sense, I defined the following categories to give shape to the model related to this specific problem using as reference the categories obtained previously with the Social Media model:

  • Cause categories: crime and insecurity (CIc) / politics, police, and justice (PPJc).
  • Effect categories: crime and insecurity (CIe) / politics, police, and justice (PPJe) / education and health services (SESe) / tourism and festivals (TFe).

Text classification was carried out with this model, and messages with no category assigned were removed. The results of all messages obtained with each category are the following:

These results reflect very different perceptions of causes and effects. For the first scenario, the effects ((50.94+46.93)% = 97.86%) sum a greater percentage than the causes ((1.64+0.5 = 2,14%), that is, people talked a lot more about the consequences of the shooting. In the second scenario, we can see that causes ((37.48+11.21)% = 48.69%) and effects ((41.94+2+0.46+6.91)% = 51.31%) are balanced, reflecting an interest in the different factors dealing with criminality in the Riviera Maya.

All results (polarities and categories) have been combined to achieve a more robust analysis, showing through bubble diagrams how the polarities of messages are divided into causes and effects. The results are the following:

We can observe that the accumulation of N+ polarity (77.11%) is largely given by the messages that speak of the consequences (75.87%), and not those referring to the causes (1.24%). Also, the concerns (negative polarities) in this scenario influence much more the effects ((14.67+75.87)% = 90.63%) that the causes ((0.64+1.24)% = 1.88%). Given the nature of the incident, this result is not surprising.

 

In this scenario, it is remarkable that the accumulation of N polarity (50.38%) comes from the same percentage of conversations classified as causes and effects (approximately 25% each).  Additionally, we can see that the conversations with negative polarities are equally distributed between causes ((24.42+9.98)% = 34.41%) and effects ((25.96+7.99)% = 33.95%), reflecting a concern both in the roots of crime and in the consequences that it will bring about in the Riviera Maya.

The previous analyses were based on all messages for polarity features and cause or effect categories. However, this global analysis hides details of what could have happened in the days of the evaluation period of the messages. For this reason, for each scenario, interactive bubble charts are provided, in order to show how the conversations’ polarities that depend on causes and effects varied daily. For better viewing these charts, please see the Appendix at the end of the post to learn how to configure them.

Scenario 1 (January 15 shooting)

Total tweets = 2102

The interactivity of this chart allows observing that on January 21, 22, and 23, the causes and the negative effects of the shooting were commented equally (bubbles moving in the middle of the chart). On the other days, there has been a tendency to talk intensely about the consequences of the incident (bubbles moving close to the Y axis).

Scenario 2 (Crime in the tourist area)

Total tweets = 651

The movement of the bubbles of the negative polarities presents a varied behavior. Before the shooting, there are days in which people talk a lot about the causes (bubbles close to X) and others days when the conversation is more focused on the consequences (bubbles close to Y). However, the shooting has accelerated the conversation regarding the consequences that this incident would bring about, both for the future of the festival and for the Riviera Maya.

 

  • Action: identifying what should be the actions to take in order to protect online reputation requires to specify what is the object (brand or product) or subject (person or organization) we are going to evaluate, as well as the objectives pursued by those who will make decisions. Both aspects are not within the scope of this post, what I want to show is how the online reputation can be measured easily with the appropriate techniques and tools.

Nevertheless, I will share in the following paragraph some conclusions based on the knowledge I acquired thanks to this study.

  1. Conclusions and suggestions

We have seen an easy and practical methodology to evaluate the online reputation of objects or subjects. The simplicity of this methodology lies in its systematic application, represented by four phases which are carried out sequentially and cyclically. The practical part consists of the use of software tools like MeaningCloud, which automatically perform complex tasks with no need to program. Besides, you can customize them to improve their performance.

On the other hand, the situations dealt globally (crime) and locally (shooting) showed results that negatively affect the Riviera Maya’s prestige as a world-class resort. Twitter interactions analyzed from December 01, 2016 reflect negative perceptions about crime’s causes and effects in the area. Negative events such as the one that occurred on January 15 amplified them.

Despite this situation, there is still the possibility of taking actions that correct and anticipate the causes and the effects of the crime that caused the rise of said negative perceptions, making use of pieces of information and knowledge like the ones we have collected in this study. In this sense, some methodological approaches such as digital marketing, social network analysis, and business intelligence can make use of the information and knowledge gathered to support the actions that are going to be taken.

In this way, as far as the Riviera Maya situation is concerned, the recommended actions are the following:

  • Digital marketing: as part of a digital communication plan, the first recommended step is to spread messages on social networks to encourage conversation on criminality. Initially, the goal is to convey that there is a will to change the negative perception brought about by crime, for which the participation of communities providing comments is required to identify the causes and specific effects of the problem. The information generated will be valuable, since it will enable to develop more accurate text classification models that help to identify the roots and the consequences of the problem.
  • Social network analysis: social listening must be combined with the information generated in the first recommended step, identifying this way the virtual communities that emerge around these conversations. These communities represent groups whose common interests (hobbies, goals, projects, lifestyles) will be valuable for designing specific corrective and preventive actions for each group.
  • Business intelligence: information analysis tools that enable to drill-down and assess the impact of each analyzed aspect should be implemented. They would help to focus the actions on those aspects that are considered relevant.

Leopoldo Martínez D.

Appendix

The chart is interactive: push the play button (the one on the left) to see the results per day in a dynamic way. Also, by clicking the speed control button (the one on the right), you can speed up or slow down the passing of days.

Set the following parameters for each diagram:

Whenever you change the visualization from one diagram to another, check the axes settings since they are not maintained.


Leave a Reply

Your email address will not be published. Required fields are marked *

*
*