help
Measurement of Hotel Service Quality Based on Online Comment Sentiment Analysis
Xuyun BAI
Faculty of Business Administration Shanxi University of Finance and
Economics Taiyuan, China
baixuy@163.com
Yan GUO Faculty of Business Administration Shanxi University of Finance and
Economics Taiyuan, China
1006727099@qq.com
Xiaotian WANG Faculty of Business Administration Shanxi University of Finance and
Economics Taiyuan, China
hahahna1974@qq.com
Abstract—To solve the problems of difficult
quantification of service quality evaluation and low
reliability of data sources, this paper proposes a hotel
service quality measurement model based on online
comment sentiment analysis and multi-attribute method.
The sentiment semantic analysis of online comments is used
as the data source of service quality measurement, and
TOPSIS multi-attribute method as the measurement
method. The effectiveness and scientific rationality of the
research method were proved by experimental analysis.
The research results not only serve as an exploration for the
application of sentiment analysis research, but also provide
a feasible technical route for the research of service quality
evaluation.
Keywords—Frame Semantics, Sentiment Analysis,
TOPSIS method, Service Quality Management
I. INTRODUCTION
The research of this paper spans two fields: text sentiment analysis and service quality management. Sentiment analysis started from the earliest text polarity classification [1], and its processing task has now been extended to the recognition of viewpoint content in sentences [2]; its research method has also developed from vocabulary-based computing to semantic-based analysis, and ideal experimental results have been obtained in many studies, such as Gangemi (2014) [3] and Li (2015) [4]. Although sentiment analysis technology can help to extract sentiment information from large-scale texts and get analysis results such as viewpoint holder, theme, viewpoint, polarity and intensity, as far as users or a user-oriented application system are concerned, the information is still too complex and needs
further comprehensive analysis and calculation to have more direct reference value for decision-making. At present, there is little research in this direction.
In the field of service quality management, due to the intangible, perishable and different characteristics of services, which are different from physical products [5], the composition and measurement methods of its evaluation elements have become difficult to study. Parasuraman (1988) [6] put forward the SERVQUAL scale, which became a classic model of service quality evaluation. However, more and more scholars questioned the practicability of this model later, and advocated establishing specific measurement models suitable for different industries according to the characteristics of the research objects, such as Albacete-Saez (2007) [7] research on tourism service quality evaluation and Bastič (2012) [8] research on hotel service. These studies have established a more diversified service quality measurement model of strong domain pertinence and explanatory power. From the perspective of research methods, previous studies mainly rely on questionnaires or interview records, which have certain limitations in terms of data scale, richness and objectivity of content. With the development of social web, some researches based on customer comment have emerged in recent years, including Song (2016) [9] and Vencovsky (2016) [10]. Online comments are the customer's conscious and spontaneous evaluation of service quality, and are the most direct reflection of customer experience. Therefore, the evaluation indexes constructed by these researches are richer and more detailed, and have obvious advantages on the data base. However, at present, these researches are still in their infancy, and the outstanding problem is the
National Social Science Foundation of China: Research on the influence of science and technology policy on the effect of enterprise innovation and
its Dynamic mechanism from the perspective of classification(20BGL019).
89
2020 Eighth International Conference on Advanced Cloud and Big Data (CBD)
978-1-6654-2313-7/20/$31.00 ©2020 IEEE DOI 10.1109/CBD51900.2020.00025
20 20
E ig
ht h
In te
rn at
io na
l C on
fe re
nc e
on A
dv an
ce d
C lo
ud a
nd B
ig D
at a
(C B
D ) |
9 78
-1 -6
65 4-
23 13
-7 /2
0/ $3
1. 00
© 20
20 IE
E E
| D
O I:
1 0.
11 09
/C B
D 51
90 0.
20 20
.0 00
25
lack of effective methods to mine sentiment information in comment texts.
Therefore, this paper attempts to combine sentiment analysis and service quality measurement, the former as the data collection and processing process, and the latter as the data calculation and analysis process. That is, first perform semantic analysis based on semantics for online comments, use the analysis results to obtain service quality evaluation attributes and evaluation data, and then use TOPSIS multi-attribute method [11] to comprehensively calculate the sentiment analysis results to realize measurement of service quality. The research model is shown in Fig. I. Considering the availability of data and the universality of practical needs, this paper sets the research scope in hotel service evaluation. The research work in this paper can be used as an exploration for the application of sentiment analysis research, and also provides a feasible technical route for the research of service quality evaluation.
Online comments
Word segmentation, part-of-speech tagging, syntactic analysis
Sentiment semantic annotation based on frame
Evaluation attribute system
Sentiment value calculation
Evaluation matrix of service quality
Calculation of comprehensive evaluation value in TOPSIS method
Service quality
evaluation D
ata collection
and processing
Fig. 1. Hotel service quality measurement model based on sentiment
analysis and TOPSIS method
II. FRAME-BASED SENTIMENT SEMANTIC ANALYSIS
A. Collection of online comments on hotel services
This paper grabs 300 customer comments for each of 5 hotels from http://hotel.elong.com/, totaling 1500, as raw data. In selection of research objects, the authors took
into account diversity and typicality, among the five hotels, hotel A is an apartment hotel, B is a five-star hotel, C is a four-star hotel, D is an budget hotel, and E is a three-star hotel.
B. Word segmentation, part-of-speech tagging and dependency syntax analysis
The purpose of word segmentation, part-of-speech tagging and syntactic analysis is to provide formal markers for sentiment word recognition and semantic role matching, which is the premise of sentiment semantic analysis. In this paper, the open source tool [12] provided by the Language Technology Platform (LTP) of the Social Computing and Information Retrieval Research Center of Harbin Institute of Technology is used, and compiled by python language to perform this task.
C. Construction of frame-based sentiment classification vocabulary
Frame is a structured category system proposed by Fillmore (1976) [13], an American linguist, to describe word meaning, sentence meaning and text meaning. With the advantages of detailed semantic classification and various types of semantic roles, frame semantic theory has attracted more and more attention and been increasingly widely applied in natural language processing and artificial intelligence. This paper classifies sentiment words by constructing a Chinese frame in the field of hotel service evaluation, and adds the description of sentiment values to the vocabulary.
Collection of sentiment vocabulary
Firstly, adjectives, verbs and nouns are extracted from the part-of-speech tagging results of hotel comments, and the sentiment words are screened according to whether they contain sentiment, feeling or evaluative semantics. Then, expand the vocabulary with the help of existing dictionaries.
Construction of frame
A group of words expressing the same cognitive scene are classified into a frame semantic class, named, and the constituent elements of the frame are determined.
Sentiment value labeling
In this paper, floating-point numbers between 0 and 1.0 are used to express sentiment values of words. Among them, 0~0.4 indicates negative evaluation, and the closer
90
to 0, the greater the intensity of negative sentiment; 0.5 means neutral evaluation, 0.6~1.0 means positive evaluation, and the closer to 1.0, the greater the intensity of positive sentiment.
Finally, the sentiment vocabulary of hotel comments constructed contains 34 frames with a total of 350 words.
The construction results are shown in Table I. Due to the space limitation, 2 words are taken for each frame as examples.
TABLE I. HOTEL COMMENTS SENTIMENT VOCABULARY
Frame Examples of words Frame Examples of words
Convenience (Convenient) (0.6), (troublesome) (0.4)
Odor characteristics (Odor) (0.4), (mildew) (0.4)
Adequacy (Enough) (0.6), (sufficient) (0.6) Sentiment experience (Satisfied) (0.7), (happy) (0.7)
Size Spacious (0.6), (large) (0.5) Social attitude evaluation (Enthusiasm) (0.6), (kindness) (0.6)
Stability (Stable) (0.6), (fixed) (0.5) Speed (Timely) (0.6), (fast) (0.6)
Quantity (Multiple) (0.6), (single) (0.4) Dryness (Moisty) (0.4), (wet) (0.4)
Development stage (Advanced) (0.6), (cutting-edge) (0.7)
Exemplary (Standardized) (0.6), (norma)l (0.6)
Richness (Rich) (0.6), (abundan)t (0.7) Suitability (Fair) (0.6), (suitable) (0.6)
Height (High) (0.5), (low) (0.5) Comfort (Comfortable) (0.6), (cozy) (0.6)
Price (Cheap) (0.6), (cost-effective) (0.8)
Particularity (Features) (0.6), (special) (0.6)
Desirability (Great) (0.7), (good) (0.6) Completeness (Thoughtful) (0.6), (complete) (0.6)
Proposal (Recommended) (0.6), (highly recommended ) (0.7)
Compatibility (Conflict) (0.4), (harmony) (0.6)
Cleanliness (Clean) (0.6), (sanitary) (0.6) Condition (Old) (0.4), (obsolete) (0.4)
Beauty and ugliness (Beautiful) (0.7), (magnificent) (0.7)
Quality (High-grade) (0.7), (high-end) (0.7)
Light or dark (Bright) (0.6), (clear) (0.6) Far or near (Next to) (0.6), (near) (0.6)
Difficulty (Easy) (0.6), (simple) (0.6) For and against (Like) (0.7), (praise) (0.6)
Tidiness (Neat) (0.6), (in order) (0.6) Worthiness (Worth) (0.6), (value) (0.6)
Quietness (Quiet) (0.6), (noisy) (0.4) Famous (Time-honored brand) (0.6), (famous) (0.7)
D. Sentiment semantic role labeling Sentiment semantic role labeling tasks include:
identifying sentiment words in comment sentences, labeling the frame to which they belong according to the sentiment classification vocabulary, and then labeling the semantic roles of the dependent components of sentiment words. Sentiment semantic roles are classified into three categories: evaluation theme, negative modification and degree modification.
Sentiment semantic role labeling mainly adopts the method based on pattern matching, that is, inputting sentences after dependency parsing, and adding semantic role markers to corresponding dependencies according to matching patterns. Semantic role collocation patterns can be summarized as follows:
emotionLU frame[SBV ATT(head) topic] [ADV-degr degree][, ADV-degr degree] [ADV-neg negative][, ADV-neg negative]
Where emotionLU is a sentiment word and its frame is matched according to the sentiment classification vocabulary. In online comments, there are often ellipsis and even single word sentences. Therefore, except for the evaluation words which are required, other components are set as optional and represented by "[]". The labeling rule is: if sentiment words dominate a subject component (SBV), the semantic role of this component is labeled as evaluation topic, as shown in the example in Fig. 2; If the sentiment words are in the position of attributive (ATT) in the attributive structure, the head is marked as the evaluation topic, as shown in the example in Fig. 3. The symbol indicates that the two cases are logical "exclusive or", that is, either SBV is true or ATT (head) is
91
true, but both cannot be true at the same time. ADV is an adverbial modifier, degr is a degree adverb, neg is a negative adverb, and their corresponding semantic role markers are degree and negative respectively.
Bed not too comfortable n d d a
SBV-Topic
Frame: Comfort
ADV-Negative ADV-Degree
Fig. 2. Examples of word segmentation, part-of-speech tagging, dependency parsing and semantic role tagging
Good (Auxiliary word) service a u v
ATT-Topic(head)
Frame: Desirability
RAD
Fig. 3. Examples of word segmentation, part-of-speech tagging, dependency parsing and semantic role tagging
In this paper, the self-developed ‘Chinese Emotional Semantic Annotation and Query System based on the FrameNet Ontology’ (Chinese Software copyright No.: 2018R11L621012) is used to realize the semantic role annotation task based on the rules above.
E. Phrase sentiment value calculation The sentiment value of a phrase represents the sentiment
polarity and intensity attribute of the whole phrase in a certain context. It takes the sentiment value of the sentiment word itself as the initial value, and then adjusts it in combination with semantic role labeling information. Firstly, for words with sentiment value of 0.5, the corresponding evaluation topics are extracted from the semantic role labeling results, and a topic adjustment scale is constructed to define the change amount of each topic word to the sentiment value of the phrase. For the change of sentiment value caused by degree modifiers, the adjustment scale of adverb sentiment value is constructed according to the degree of each adverb's effect on sentiment intensity. For example, the adjustment amount of the adverb "very" is ±0.1, indicating that the sentiment value of the phrase is enhanced by 0.1 compared with the sentiment value of the original sentiment word, specifically which for the commendatory word is increased by 0.1, while that for the derogatory word
is decreased by 0.1. The change of sentiment value caused by negative modifiers is regulated by the rule of (1- original value). At the same time, the adjusted sentiment value cannot be greater than 1 or less than 0. Fig. 4 below summarizes the sentiment value algorithm:
X=X0
X=X+a
X=X+b
X=0
X=1-X
X=0.5
With degree modification?
X>1
X<0
With negative modification?
X=1
Y
N
N
Y
Y
N
Y
N
Y
N
is the keyword adjustment amount.
is the adverb adjustment amount.
is the sentiment value of sentiment words.
Fig. 4. Phrase sentiment value algorithm
III. CONSTRUCTION OF HOTEL SERVICE QUALITY EVALUATION ATTRIBUTE SYSTEM
The attribute of hotel service quality evaluation is determined according to the results of sentiment semantic analysis. Firstly, the evaluation attributes are preliminarily determined according to the frame of sentiment words. Then, the evaluation topics are clustered, that is, the vocabulary semantic similarity calculation software of HowNet (download address: http://www.keenage.com/html/c_index.html) is used to classify the topic words with similarity greater than or equal to 0.8 into one class, and then the underlying classes are further summed up by hierarchical clustering method, and the clustering results are adjusted to classify the unrecognized words into appropriate categories. Finally, the hotel service quality evaluation attribute system is formed, which is divided into 7 aspects and 49 attributes, as shown in Table II.
TABLE II. HOTEL SERVICE EVALUATION ATTRIBUTES
Category Attribute
92
Hotel overall F1: Advantages and disadvantages of hotel as a whole (high-end, luxury, etc.). F2: Aging degree of hotel as a whole. F3: Whether it is suitable for certain travel intentions (business or travel, etc.).
Guest room
Guest room overall
F4: Guest room overall satisfaction. F5: Aesthetic feeling in guest room layout. F6: Guest room area. F7: Guest room living condition. F8: Guest room sanitation. F9: Guest room design satisfaction. F10: Guest room design height. F11: Guest room design novelty. F12: Guest room decoration aging degree.
Facilities
F13: Whether windows are satisfactory (including whether there are windows or not, orientation, etc.). F14: Size of windows. F15: Whether facilities are satisfactory. F16: Whether facilities are complete. F17: Aging degree of facilities. F18: Aging degree of furniture. F19: Whether there is Wi-Fi. F20: The sanitary condition of fabrics in rooms (sheets, towels, etc.). F21: The performance of air conditioners in rooms. F22: Whether the size of beds is satisfactory. F23: The comfort degree of beds.
Bathroom
F24 Sanitary condition of bathroom. F25 Whether bathroom is convenient to use. F26 Performance of bathroom facilities. F27 Aging degree of bathroom facilities. F28 Aging degree of toiletries.
Geographic location
F29: Whether the geographical location of is satisfactory on the whole. F30: Whether the geographical location of is convenient. F31: Whether the distance to the travel destination is satisfactory. F32: Whether the traffic conditions around are satisfactory. F33: Whether transportation is convenient. F34: Whether the landscape around is satisfactory
Internal environment
F35: Whether the internal environment of hotel is satisfactory. F36: Whether the internal environment of hotel is quiet. F37: Whether the internal environment of hotel is beautiful. F38: Sanitation of the internal environment of the hotel.
Services
F39: Whether hotel service is generally satisfactory. F40: Service staff's attitude in general. F41: Whether check-in and check-out service is satisfactory. F42: Check-in and check-out service staff's attitude.
Catering
F43: Whether the catering service of hotel is satisfactory on the whole. F44: Whether the breakfast service of hotel is satisfactory (including the availability and quality of breakfast, etc.). F45: Whether the breakfast of hotel is rich. F46: How is the taste of breakfast.
Price F47: Whether the price is satisfactory on the whole. F48: Whether the price is high or low. F49: Whether it is worth the price.
IV. HOTEL SERVICE QUALITY MEASUREMENT BASED ON TOPSIS MODEL
TOPSIS multi-attribute method (Technique for Order Preference by Similarity to an Ideal Solution) is a sequential optimization technique based on the similarity of ideal objectives proposed by Hwang and Yoon in 1981 [11]. It is very suitable for solving the quality evaluation problems of multiple evaluation objects and multi-attribute. Therefore, this paper uses this model to measure hotel service quality. The specific steps are as follows:
The first step is to build an online comment data matrix nmija )(A . Where aij represents the sentiment value of
the i-th evaluation object on the j-th evaluation attribute, m and n are the number of evaluation objects and evaluation attributes respectively. The sentiment value of the evaluation object on each evaluation attribute is expressed by the average of the sentiment values of all the customer comments of the object on this attribute, namely
Where P is the number of comment phrases of the i-th
evaluation object containing the j-th attribute, kR represents the sentiment value of the comments phrase containing the j- th attribute.
In the second step, the original sentiment value data is dimensionless processed to obtain the dimensionless
standardized matrix nmij rA )(
, where
93
The third step is to set the evaluation attribute weight. The weight of each attribute is calculated by the number of its occurrences. The process is as follows:
wj is the weight of the j-th attribute; Pj is the number of times that the j-th attribute appears in product comments. The weight of an attribute is calculated by the proportion of the number of occurrences of the attribute in the sum of the number of occurrences of all attributes.
The fourth step is to calculate the weighted normalization matrix
Where, wj is the weight of attribute j, and rij is the evaluation value after dimensionless processing.
The fifth step is to determine the positive and negative ideal solutions of the evaluation attributes. Use matrix B to get the maximum and minimum values of the sentiment values of all attributes in all comments, that is, positive ideal solution (B+) and negative ideal solution (B-).
The sixth step is to calculate the distance between the evaluation object and the positive and negative ideal solutions
(i=1,2, ,m; j=1,2, ,n) (8)
(i=1,2, ,m; j=1,2, ,n) (9)
The seventh step is to calculate the proximity Si between the evaluation object and the optimal scheme.
(0 ≤ Si ≤ 1, i=1,2, ,m) (10)
Where, Si is between 0 and 1, and the larger the value, the closer the service quality evaluation index of the evaluation object is to the ideal value, and the higher the service quality; on the contrary, the smaller the value, the worse the service quality evaluation of the evaluated object is.
V. EXPERIMENTAL ANALYSIS
A. Data sources In this paper, the authors used the comments data in
Section II to eliminate some invalid texts, including too few comment phrases corresponding to sentiment words and phrases without sentiment semantics. Finally, 1014 effective comments phrases were obtained.
B. 5.2 Experimental results The evaluation attribute weight calculation result is:
W= (0.0059, 0.0099, 0.0069, 0.0089, 0.0059, 0.0582, 0.0276, 0.0878, 0.0108, 0.0059, 0.0059, 0.0059, 0.0069, 0.0069, 0.0197, 0.0108, 0.0394, 0.0059, 0.0089, 0.0059, 0.0069, 0.0059, 0.0099, 0.0069, 0.0069, 0.0069, 0.0046, 0.0059, 0.0779, 0.0365, 0.071, 0.0069, 0.1026, 0.0059, 0.0444, 0.0158, 0.0089, 0.0069, 0.0457, 0.0434, 0.0059, 0.0069, 0.0118, 0.0069, 0.0138, 0.0059, 0.0178, 0.0444, 0.0079)
The attribute with the highest weight above is F33, that is,
the convenience of traffic conditions, which indicates that it is very important whether the traffic is convenient from the customer's point of view. The other two attributes related to geographical location: F29 (Whether the geographical location of is satisfactory on the whole) and F31 (Whether the distance to the travel destination is satisfactory) are also ranked high in weight, ranking third and fourth respectively. It can be seen that the convenience of traveling from the hotel to the destination will greatly affect the customer's evaluation of the hotel. F8 (Guest room sanitation) ranks second in weight, which reminds us that cleaning should be one of the key points of hotel management.
Table III shows the weighted standardized evaluation values. Due to the space limitation, only the evaluation values corresponding to the top 10 attributes with weight rearrangement are listed here.
TABLE III. WEIGHTED STANDARDIZED EVALUATION VALUE a
Sr. No.
A B C D E
F33 0.0429- 0.0466 0.0448 0.0467 0.0481+
F8 0.0367- 0.0393 0.0397 0.041+ 0.0395 F29 0.037+ 0.0351 0.0321- 0.0349 0.0349
F31 0.0313 0.0321 0.0311- 0.0324+ 0.0319 F6 0.0294 0.0272 0.0243 0.0306+ 0.0162-
F39 0.0225+ 0.0199 0.019- 0.0212 0.0193 F35 0.0203 0.022+ 0.0184- 0.0188 0.0196
F48 0.024+ 0.0228 0.0189 0.0174 0.0146- F40 0.0183- 0.0203+ 0.0192 0.0199 0.0192
F17 0.0203+ 0.0135- 0.0193 0.0193 0.0146
a. Note: The value marked with ‘+' in the upper right corner is a positive ideal solution, and the value marked with ‘-' is a negative ideal solution.
It can be seen from Table 3 that Hotel A occupies a positive ideal solution in four attributes, namely F29, F39, F48 and F17, indicating that the hotel has been highly evaluated in terms of geographical location, service, price
94
and internal facilities, and is the benchmark among the five hotels in these four aspects. Hotel C does not occupy any positive ideal solution in 10 attributes with high weights, but has 4 attributes with negative ideal solutions, so the overall evaluation of the hotel is expected to be relatively poor.
The distances D+ and D- between each evaluation object and the positive and negative ideal solutions are calculated according to formulas (8) and (9), and then the relative proximity S between each evaluation object and the ideal solution is calculated according to formula (11), and ranked according to the S value from large to small. The results are shown in Table IV:
TABLE IV. PROXIMITY AND RANKING RESULTS
Hotel D+ D- S Rank-ing A 0.009416837 0.020465175 0.684866034 1 B 0.01112859 0.017460582 0.610741091 2 C 0.013805882 0.013200149 0.488785219 4 D 0.011680109 0.018133358 0.608227092 3 E 0.020737537 0.009221596 0.307805841 5
According to the ranking results of the proximity to the ideal solution, the ranking order of the five evaluation objects is A>B>D>C>E, which shows that according to the evaluation and feedback of customers, the overall service quality of Hotel A is the highest, while that of Hotel E is the lowest. Surprisingly, although Hotel E is a three-star hotel, its ranking result is not as good as that of the budget hotel D. The customer evaluation reflects the satisfaction degree of the customer's actual experience comparing to their expectations after staying in the hotel. Although Hotel E is superior to Hotel D, and should be better than Hotel D in hardware and environment, Hotel E is lower than Hotel D in customer satisfaction compared with the price paid by customers and expectation before check-in.
From the above experimental results, it can be seen that, based on the sentiment semantic analysis of customer comments and TOPSIS attribute method, we can not only obtain the specific evaluation results of the evaluation object on a certain attribute, but also get comprehensive ranking information, which can provide reference for hotel management decision-making from multiple levels and angles.
VI. CONCLUSION In this paper, sentiment semantic analysis and TOPSIS
multi-attribute method are combined to measure the hotel service quality effectively. The contributions of the research include: taking online comments as the data source and sentiment analysis as the data processing method, this paper
can reflect the customer experience more truly and carefully than the questionnaires and interviews used in previous research, thus ensuring the validity and practicability of the measurement results; In the sentiment analysis method, this paper adopts the frame-based semantic analysis technology to produce refined semantic classification and comprehensive sentiment information; With TOPSIS multi- attribute method and the number of comments as the weight, the hotel service quality is scientifically and effectively measured.
The further research work is to expand the scope of research fields and experimental objects, so that the method proposed in this paper has wider application value.
REFERENCES
[1] Pang B, Lee L, Vaithyanathan S, “Thumbs up? Sentiment classification using machine natural learning techniques,” in proceedings of the 2002 Conference on Empirical Methods in Natural Language Processing, 2002, pp.79-86.
[2] Chelaru S, Altingovde I S, Siersdorfer S, “Analyzing, detecting, and exploiting sentiment in web queries,” in ACM Transactions on the Web, 2013, 8(1), pp.1-33.
[3] Gangemi A, Presutti V, Recupero D R, “Frame-based detection of opinion holders and topics: A Model and a Tool,” in IEEE Computation Intelligence Magazine, 2014, 9(1), pp. 20-30.
[4] Li S, Huang L, Wang J, Zhou G, “Semi-stacking for Semi-supervised sentiment classification,” in proceedings of the 53rd Annual Meeting of the Association for Computational Linguistics and the 7th International Joint Conference on Natural Language Processing, 2015, pp.27-31.
[5] Zeithaml V A, Parasuraman A, Berry L L, “Problems and strategies in services marketing,” Journal of Marketing, 1985, 49(2), pp. 33-46.
[6] Parasuraman A, Zeithaml V A, Berry L L, “SERVQUAL, A multiple- Item scale for measuring consumer of perception service quality,” Journal of Retailing, 1988, 64(1), pp.12-40.
[7] Albacete-Saez C A, Fuentes-Fuentes M M, Llorens-Montes F J., “Service quality measurement in rural accommodation,” Annals of Tourism Research, 2007, 34(1), pp.45-65.
[8] Bastič M, Gojčič S., “Measurement scale for eco-component of hotel service quality,” International Journal of Hospitality Management, 2012, 31(3), pp.1012-1020.
[9] Song B, Lee C, Yoon B., “Diagnosing service quality using customer reviews: an index approach based on sentiment and gap analyses,” Serviece Business, 2016, 10(4), pp.775-798.
[10] Vencovsky F, Bruckner T, Sperkova L., “Customer feedback analysis: case of e-banking service,” in proceedings of the 3rd European Conference on Social Media, 2016, pp.404-414.
[11] Hwang C L, Yoon K., “Multiple attribute decision making: methods and applications,” New York: Springer-Verlag, 1981.
[12] Che W, Li Z, Liu T, “LTP: A Chinese language technology platform,” in proceedings of the Coling 2010: 13-16.
[13] Fillmore C J., “Frame semantics and the nature of language,” Annals of the NY Academy of Sciences, 1976, 280, pp.20-32.
95