Analyzing Online Car Reviews Using Text Mining

Data was collected using the scraping tool “ParseHub.” This is a useful instrument when dealing with information of any kind. It can be adapted to any website, and scholars can extract any piece of information, be it a text or an image (e.g., Figure 1). In this study, units such as title, model of vehicle, best features, worst features, ratings for 8 different features, and total ratings were collected. The process of scraping is illustrated in Figure 1.

The results can be saved in JSON or CSV format. In this study, they were saved to CSV format, slightly corrected in an Excel program, and then changed to XSLX format for further usage. The output appears as shown in Figure 2.

All reviews were divided into two groups: satisfied and unsatisfied. The overall rating by consumers was used as the basis for this division. The average mean of overall ratings thus became a separating point for all three car samples and the next condition was set as follows:

where Car #1 is a Hyundai Elantra, Car #2 is a Honda Civic, and Car #3 is a Ford Focus. GD is a variable for group diversity where GD = 1 relates to the satisfied group and GD = 0 relates to the unsatisfied group. The result of the data split is presented in Table 1.

For further analysis, data was input into the R program using the “xlsx” package. This enables the R user to read, write, and format Excel files. Best features, worst features, satisfied and unsatisfied variables, reviewers’ id, name of automobile brand, and car id relative to automobile brand were included as input. Part of the input data is displayed in Figure 3.

In the output we acquired 539 reviews, each of which referred to either the satisfied group or unsatisfied group in accordance with the sat/unsat variable obtained from the average mean of the overall rating.

After the data were superficially arranged, R program tools were used to process the data and divide the reviews into four different groups: satisfied best features, satisfied worst features, unsatisfied best features, and unsatisfied worst features. Such segregation can be deciphered as meaning that, even though consumers may be satisfied or unsatisfied, they still identify some best and worst features of the car they are reviewing. An example of the output is presented in Figure 4, where rows are organized randomly to increase clarity.

To make it possible to count words and identify their co-occurrence in reviews, text must first undergo the cleaning process. Some features of text are thus removed, such as numbers, white spaces, punctuation, and common English words that have no semantic meaning. As well as converting text to lower case, fixing contractions and text stemming is also essential to obtain accurate and valuable data. We therefore received all the words occurred in all 1078 reviews, which totaled 2299 words. A sample of the output is presented in Figure 5.

For further data analysis, the final step was to combine our primary database and database, which consisted of occurring words. Thus, with the help of R tools, a master table was created, as shown in Figure 6.

Figure 7 shows the best features of three cars. Words with the highest co-occurrence are represented in this word cloud, with the most frequent and important words located in the center and the least frequent words located on the edges. Hence, the closer the words are to edges, the less frequent they are. In the case of Hyundai, “seat” is the most frequent word, followed by “interior,” then “style,” and then the rest. In the case of Honda, the most frequent words are, in order, “mpg,” “gas,” “seat,” “comfort,” “mileage,” “dash,” “control,” “display,” “smooth,” “steering,” “wheel,” “econ,” “fun,” and so on. In contrast with Hyundai, where the main advantage was design and style, consumers mostly emphasize characteristics related to value, technology, and movement on a road. In the case of Ford, the most frequent words are “handles,” “seat,” “interior,” “system,” “sync,” “style,” “comfort,” “gas,” “transmission,” “exterior,” “mileage,” and so on.

In the case of Hyundai, after filtration only 14 words remain. As shown in the barplot in Figure 8, there are many words relating to appearance. These are “interior” (24), “style” (19), “exterior” (13), “look” (13), and “design” (10). The interpretation of this result is that consumers mostly liked the car design. The most frequent word is “seat,” which occurs 43 times. Because this word has such high frequency, association analysis was conducted to determine its significance. We performed correlation analysis on the most frequent word, as shown in Figure 9. For example, in the case of Hyundai, the term “seat” has a high correlation with words such as “position,” “front,” and “back.” Hence, we can assume that this word refers to the convenience and comfort consumers felt when they sat in a Hyundai Elantra. Consistent with this interpretation, words such as “back,” “comfort,” and “rear” might also refer to comfort, which was one of the best features for consumers. Similarly, the occurrence of words such as “mpg” and “gas” means that consumers were satisfied with Hyundai’s fuel consumption. For the word “control,” the most closely associated word was “steering” with a correlation of 0.78. This means consumers were likely to be satisfied with their control over the movement of a vehicle.

In the case of Honda, the first two words are “mpg” and “gas” which means this car has very low fuel consumption. Additionally, for some consumers, seats seem to be very comfortable. The rest of the words are related to the dashboard and technological features such as “dash,” “control,” “display,” “steering,” “system,” “bluetooth,” and “econ.” The word “steering” refers more to technology than holding the road because the correlated words were “wheel,” “control,” “electronic assist,” and “dash.” The word “econ” was correlated with words such as “mode” and “feature.” This is explained by the fact that the Honda Civic has an econ button as a special function, which has become one of its most favored features.

In the case of Ford, the most frequent word is “handles,” which occurred 48 times. Because it is quite difficult to interpret this word, an association analysis was conducted as shown in Figure 10. We can assume that the word “handles” does not refer to the means by which a thing is held, carried, or controlled, but how easily a car is to handle on a road. Such words as “turn,” “directions,” “balance,” “quiet,” “turn,” and others can help to precisely interpret the meaning of this word. Another frequent word was “sync,” which are correlated with some words such as voice, system, phone, ipod, control, navigation, and so on.

The same analysis was then conducted using reviews that contained the worst features of three brands. For a Hyundai Elantra, the most frequent word is “mpg,” which occurred 28 times, as shown in Figure 11. Although “mpg” also occurred in the results for best features, it is not impossible for the same term to appear in worst features. Here, we can assume that many consumers were not satisfied with fuel consumption and that these consumers outnumber those who were satisfied.

Moreover, if we consider the correlation analysis for “mpg” in Figure 12, we can see that there are highly correlated words such as “show,” “computer,” “onboard,” and “display.” Therefore, we assume there might be some problem related to displaying the mpg on the onboard computer. This hypothesis was checked manually, and it was found that many consumers were complaining about an incorrect mpg display.

Consistent with this finding, the correlation for the word “gas” yielded a similar result as shown in Figure 13, so we can assume that the words “estimates” and “misleading” are referring to the same problem.

The barplot for worst features shows that consumers were unsatisfied with spare tire (“spare” and “tire” were the most highly correlated with = 0.89), noise on the road, fog lights, trunk, mpg efficiency, as well as the mpg display and seats. As mentioned previously, “mpg,” “gas,” “fuel,” as well as “seat,” occurred inconsistently in the results for best features. Such phenomena could be accounted for by the differing preferences of every individual. Furthermore, based on the proportion of words for both groups, mpg and mileage are more likely to be considered poor rather than superior features because the sum of occurrences in best features is 36; in worst features it is 61.

In the case of Honda, we can see that one of the most frequent words in terms of worst features is “interior,” as shown in Figure 14.

The correlation graphic shows that this is highly correlated with “cheap,” as shown in Figure 15. Therefore, we can assume that some customers did not like the quality and appearance of their interior, viewing it as a drawback rather than an advantage. Additionally, “fabric” is correlated with “interior.” Although the correlation is quite low, we can still assume that reviewers were unsatisfied with the material of their interior. Furthermore, some consumers were not satisfied with back or front seats. Moreover, fog lights, mirrors, and noise on roads became some of the worst features in the Honda Civic as shown in Figure 14.

In the case of Ford, the worst features were “transmission,” “seat,” “back,” “control,” “fix,” “issue,” “shift,” “rear,” “wheel,” “system,” and so on. The most frequent word “transmission” occurred 57 times, comparatively larger than other terms. To understand the transmission flaw in the Ford Focus, an association graph was built, as shown in Figure 16.

Among the terms highly correlated with “transmission” were “severe,” “grinding,” “crunching,” and “bucking.” Therefore, we can assume there is a problem with the transmission, as it is perceived as making strange sounds and being inconvenient to use. In addition, the terms “issue,” “fix,” “manual,” “problem,” and “shift” also correlated with “transmission,” which means that it is probably the most significant problem with the Ford Focus. Ford also seems to have problems in terms of technology. For instance, the terms “control” and “wheel” were highly correlated with “steering,” “device,” “equipment,” “aux,” “cruise,” “dashboard,” and other words, which can be interpreted as Ford exhibiting a deficiency in equipment. There are also consumers who are certainly not satisfied with the seats and space in a cab, both front and back (rear).

Based on the results, we can assume that the biggest strength of the Hyundai Elantra car is its design. This is supported by the fact that the most frequent words are related to car appearance. These include “interior,” “style,” exterior,” “look,” and “design.” The worst features for Hyundai appear to be gas consumption and some problems with technology, such as the mpg display on the onboard computer and problems with a spare tire. In contrast to the Hyundai car, the biggest strengths of the Honda car are low gas consumption, dashboard, controls on the steering wheel, and the “econ” mode, which improves fuel efficiency. The worst feature for Honda appears to be its interior, which reviewers emphasized as cheap. Furthermore, they were unsatisfied with the material it was made of. In the case of Ford, the biggest strength appeared to be manipulation of the car. This is consistent with the high frequency of the word “handles.” The, other best features for Ford were the interior and exterior, noiselessness during the ride, and the “Ford Sync” system which allowed users to control automotive functions using their voice. The biggest disadvantage for Ford was found to be transmission. This was supported by the high frequency of the word “transmission” and other frequent yet negative words, such as “issue,” “fix,” “manual,” “problem,” and “shift.” Another important disadvantage relates to technology, specifically a problem with the controls on the steering wheel. In addition, in all three cases one of the most frequent words was “seat,” which appeared in both “best features” and “worst features” categories, suggesting that reviewers are divided in their opinions. Hence, it can be concluded that it is difficult for all three car brands to find favor in the eyes of all consumers.

As shown in Table 3, the highest frequency was for terms related to comfort and interior features in Hyundai, comfort and technological features in Honda, and comfort in Ford (Criteria: F ≥ 1). Therefore, reviewers were mostly interested in these features and discussed these most heatedly. These can now be scrutinized more closely for each case.

(1) Hyundai

If we compare the frequency of terms for interior and the average rating score of consumers for interior we see that terms related to interior appear more often in the satisfied group and in best features because the score for interior is quite high. Therefore, the Hyundai company has won the favor of consumers in respect to its interior. The comfort score is 3.97, which is neither high nor low. This suggests there might be some factors reviewers were not satisfied with. Thus, a more precise analysis is needed. Another feature worth considering is the exterior feature. Compared to other automobile brands, the frequency of terms related to the exterior is significantly higher than for Honda and Ford. It can therefore be assumed that the exterior is also the strongest feature for Hyundai. Its score is highest among the scores for all features and the likelihood that terms related to an exterior would mostly occur in the satisfied group and best features is very high.

(2) Honda

Comparing the frequencies and scores for comfort and technology, it is very likely that terms related to these will occur mostly in the satisfied group and best features because the score for both groups is pretty high. Furthermore, the frequency and score for technology is the highest among all three automobile brands, which can be interpreted as Honda being a technological leader.

(3) Ford case

Although the frequency of terms related to comfort is high, the score is quite low. Although it is not the lowest score compared to other features, the result suggests that such terms would appear in both satisfied and unsatisfied groups and in both the best and worst features groups. Furthermore, reliability is also worth mentioning, because the frequency of terms related to this feature is much higher than for Hyundai and Honda. With a very low score for reliability, we can assume that terms will mostly appear in worst features for both satisfied and unsatisfied groups.

In this section, we compare the reviews of both groups and find terms whose influence is greater than others. We also compare the differences between satisfied and unsatisfied groups of reviewers. What, therefore, are the frequency and ratio of terms for eight different features between satisfied and unsatisfied groups and what are the implications of this?

Based on the results of this research, the following propositions can be stated. Firstly, among the three car brands, the Hyundai Elantra car has the best marks in relation to “interior” and “exterior,” but, in terms of the interior, there is a problem with a spare tire that needs to be solved. A few people were also unsatisfied with gas consumption, so it would be better for engineers from Hyundai to improve the mpg index. Furthermore, Hyundai has problems in terms of technology, one of which is incorrect mpg displays. Moreover, despite consumers’ satisfaction with comfort, there was a problem with noise on roads. This seems to be the main reason for a relatively low score for comfort. Hence, Hyundai should reconsider the value and technological particularities of the Elantra car to make it more competitive on the market. Secondly, among the three car brands, the Honda Civic received positive feedback for all features and was found to be best in terms of value and technology. It has the best mpg index and the best technological equipment compared to Hyundai and Ford. Despite satisfaction with all features, Honda engineers should pay attention to the interior, because many consumers criticized it for its cheapness. In addition, Honda should consider the issue of comfort, because back seats were also found to be a weak point. Thirdly, the Ford car was evaluated very poorly regarding all features with the exception of the exterior, where it received a high score. However, according to the results, this was not the most discussed topic among reviewers. Among the 24 most frequent terms, negative terms were in the majority. These were related to the topic of reliability, where reviewers severely criticized transmission and found this to be the biggest problem in the Ford automobile. Apart from problems with transmission, most consumers were quite unsatisfied with both back and front seats. The only feature reviewers were truly pleased with, according to the results, was manipulation of the car. In previous research, reliability was found to be one of the most significant factors for buyers. Therefore, looking at the poor evaluation of this car, reviewers were greatly disappointed with its reliability and, for this reason, the scores for other features were slightly biased. Hence, Ford marketers and engineers must completely reconsider and reassess their car from all sides, starting with the reliability feature.

In addition, the results for several features, such as safety, were unclear and ambiguous, thereby making interpretation difficult. This can be explained by the lack of terms chosen for the analysis. To fill such blind spots, a more extended analysis is needed.