Response Generation by Jointly Modeling Personalized Linguistic Styles and Emotions
Natural language generation (NLG) has been an essential technique for various applications, like XiaoIce and Siri, and engaged increasing attention recently. To improve the user experience, several emotion-aware NLG methods have been developed to generate responses coherent with a pre-designated emotion (e.g., the positive or negative). Nevertheless, existing methods cannot generate personalized responses as they frequently overlook the personalized linguistic style. Apparently, different human responsers tend to have different linguistic styles. Inspired by this, in this work, we focus on a novel research theme of personalized emotion-aware NLG ( PENLG ), whereby the generated responses should be coherent with the linguistic style of a pre-designated responser and emotion. In particular, we study PENLG under a scenario of generating personalized emotion-aware response for social media post. Yet it faces certain research challenges: (1) the user linguistic styles are implicit and complex by nature, and hence it is hard to learn their representations; and (2) linguistic styles and emotions are usually expressed in different manners in a response, and thus how to convey them properly in the generated responses is not easy. Toward this end, we present a novel scheme of PENLG, named CRobot, which consists of a personalized emotion-aware response generator and two discriminators, i.e., general discriminator and personalized emotion-aware discriminator . To be more specific, the post-based and avatar-based user linguistic style modeling methods are incorporated into the encoder-decoder–based generator, while the discriminators are devised to ensure that the generated response is fluent and consistent with both the emotion and the linguistic style of the user. Different from the traditional adversarial networks, we embed adversarial learning under the umbrella of reinforcement learning. In this way, the response generation problem can be tackled by the generator taking a sequence of actions on selecting the proper word of each timestep for output. To justify our model, we construct a large-scale response generation dataset based on Twitter, consisting of 6,763 tweets with a corresponding 1,461,713 response created by 153,664 users. Extensive experiments demonstrate that CRobot surpasses the state-of-the-art baselines regarding both subjective and objective evaluation.
- Research Article
- 10.61707/n2bdvb28
- Jul 19, 2024
- International Journal of Religion
Due to the various and multiple linguistic communication styles and methods, context becomes of paramount importance in studying the meaning carried by the text of these styles and investigating it, as it is based primarily on a direct relationship between the addresser and the addressee. However, style is the method of writing, or the way of choosing and composing words to express meanings with the aim of clarification and influence. Based on the concept of the sentence and its two components, and its division into nominal and verbal, the science of meanings emerged divided into two chapters: the chapter of news and the chapter of construction. Thus, the grammarians' attention to the context was of great importance to reach the exact meaning of each of these styles, especially they may deviate from what they were established for, and require other completely different meanings and implications, as was previously clarified when discussing the concept of contextual necessity.
- Research Article
- 10.3390/nu18010044
- Dec 22, 2025
- Nutrients
This study investigated the relationship between time spent on social media and eating behaviours among young Australian adults. It also examined the types of content discussed and linguistic styles used by health and nutrition content creators on Instagram. Young adults (aged 18-30 years) who reported viewing social media for nutrition or health content were recruited to complete a self-administered, cross-sectional survey. Data on demographics, time spent on Instagram and TikTok, health content creators viewed, and responses to the Scale of Effects of Social Media on Eating Behaviours (SESMEB) were collected. Associations between time spent on Instagram and TikTok and SESMEB scores were analysed. Inductive content and thematic analysis were conducted on health-related posts from Instagram accounts viewed by study participants. From the 57 participants who completed the demographic survey, 42 participants completed the full study including the SESMEB survey. There was no significant association between SESMEB score and time spent on Instagram (p = 0.38) or TikTok (p = 0.40). A total of 1420 Instagram posts from 71 distinct content creators were analysed. Health and fitness product endorsements or advertisements (56.3%), predominantly posted by laypersons (55.3%), were the most common type of post in the sample. The most common communication style was 'expert advice' (47.9%), with 'informal language' (85.9%) as the dominant linguistic style. Results from thematic analysis suggest health and nutrition information on social media is often presented to consumers in emotionally charged, stylised, or contradictory ways and requires users to sift through conflicting messages, aesthetics, and ideologies to construct their own understanding of health. This study suggests that young adults are primarily exposed to health and fitness product promotions from unqualified content creators on social media. Dietitians and nutrition professionals may need to consider adopting specific linguistic and communication styles to enhance the dissemination and engagement of credible nutrition information online. These findings have implications for improving digital health literacy and strengthening the impact of evidence-based nutrition messaging in digital environments.
- Research Article
- 10.2478/amns-2025-0319
- Jan 1, 2025
- Applied Mathematics and Nonlinear Sciences
A study of text mining of Japanese academic articles to analyze the linguistic style changes in Japanese academic articles. A dataset of Japanese academic articles is constructed and the text data is preprocessed. The linguistic feature model of Japanese academic articles is constructed, and the linguistic features of vocabulary, sentences, and other measures are extracted based on the original text data for linguistic style analysis. By analyzing the changes in word length, sentence length, and lexical richness of Japanese academic articles between 1981 and 2020, the linguistic style evolution of Japanese academic articles during this period is explored.The average word length of Japanese academic articles between 1981 and 2020 is in the range of [1.8329, 1.9507], and the word length dispersion is in the interval of [0.338, 0.362]. The frequency of monosyllabic and disyllabic words has shown a slow decreasing trend, but they still remain the most frequently used word classes in Japanese academic articles. The average sentence length increased from 43.58 to 49.27, which is associated with an increase in text complexity and formality. The percentage of sentence lengths of 1~15 and 16~30 is around 50%. The proportion of sentences with length >45 is generally on the rise, and the linguistic style of Japanese academic articles tends to be more and more standardized and rigorous.The vocabulary density of Japanese academic articles during the 20-year period is in the range of 0.7936-0.8711, and the type-case ratios are in the range of 6.9418-35.8726.The vocabulary of Japanese academic articles in the period of 2001-2005 is the most abundant.
- Research Article
65
- 10.1177/0956797619894557
- Dec 26, 2019
- Psychological Science
This research demonstrates that linguistic similarity predicts network-tie formation and that friends exhibit linguistic convergence over time. In Study 1, we analyzed the linguistic styles and the emerging social network of a complete cohort of 285 students. In Study 2, we analyzed a large-scale data set of online reviews. In both studies, we collected data in two waves to examine changes in both social networks and linguistic styles. Using the Linguistic Inquiry and Word Count (LIWC) framework, we analyzed the text of students’ essays and of 1.7 million reviews by 159,651 Yelp reviewers. Consistent with our theory, results showed that similarity in linguistic style corresponded to a higher likelihood of friendship formation and persistence and that friendship ties, in turn, corresponded to a convergence in linguistic style. We discuss the implications of the coevolution of linguistic styles and social networks, which contribute to the formation of relational echo chambers.
- Research Article
2
- 10.2139/ssrn.3131715
- Mar 7, 2018
- SSRN Electronic Journal
This paper demonstrates that linguistic similarity predicts network tie formation and that friends exhibit linguistic convergence over time. Study 1 analyzes the linguistic styles and the emerging friendship network in a complete cohort of 285 students. Study 2 analyzes a large-scale dataset of online reviews. Across both studies, we collected data in two waves to examine changes in both friendship networks and linguistic styles. Using the LIWC linguistic framework, we analyze the text of students’ essays and of 1.7 million reviews by 159,651 Yelp reviewers. We find that similarity in linguistic style corresponds to higher likelihood of friendship formation and persistence, and that friendship ties, in turn, correspond with a convergence in linguistic style. We discuss the implications of the co-evolution of linguistic styles and social networks, which contribute to the formation of relational echo chambers.
- Research Article
70
- 10.1287/isre.2022.1107
- Feb 16, 2022
- Information Systems Research
Firms of all sizes are “joining the conversation” on social media platforms and increasingly trademarking hashtags related to their products and brands. This added effort to protect intellectual property and its impact on social media engagement have not been investigated in the literature. In this study, we find that trademarking hashtags plays a pivotal role in increasing social media audience engagement and information dissemination. More importantly, this positive effect is stronger for firms with fewer Twitter followers. Digging deeper into the underlying mechanisms, we find that trademarking hashtags makes composing tweets with certain linguistic styles more critical: It can amplify the positive effects of trademarking hashtags on social media audience engagement. Our findings highlight important managerial implications of trademarking hashtags. First of all, we examine whether trademarking a hashtag helps or hurts a firm in terms of its social media audience engagement. Further, we show, to maximize the effectiveness of trademarking hashtags, how firms should develop the right social media engagement strategies by taking specific communication and linguistic styles into account. Our results provide useful insights to firms in understanding the key benefits of signaling through trademarking hashtags on social media engagement.
- Research Article
185
- 10.1162/coli_a_00063
- Sep 1, 2011
- Computational Linguistics
Recent work in natural language generation has begun to take linguistic variation into account, developing algorithms that are capable of modifying the system's linguistic style based either on the user's linguistic style or other factors, such as personality or politeness. While stylistic control has traditionally relied on handcrafted rules, statistical methods are likely to be needed for generation systems to scale to the production of the large range of variation observed in human dialogues. Previous work on statistical natural language generation (SNLG) has shown that the grammaticality and naturalness of generated utterances can be optimized from data; however these data-driven methods have not been shown to produce stylistic variation that is perceived by humans in the way that the system intended. This paper describes Personage, a highly parameterizable language generator whose parameters are based on psychological findings about the linguistic reflexes of personality. We present a novel SNLG method which uses parameter estimation models trained on personality-annotated data to predict the generation decisions required to convey any combination of scalar values along the five main dimensions of personality. A human evaluation shows that parameter estimation models produce recognizable stylistic variation along multiple dimensions, on a continuous scale, and without the computational cost incurred by overgeneration techniques.
- Research Article
35
- 10.1111/clr.12201
- Jun 10, 2013
- Clinical Oral Implants Research
While extensive references are present in the literature dealing with the correlation between subjective and objective evaluation of tooth shade, there is a lack of information on this correlation regarding the soft tissue color. The purpose of this experimental study was to verify whether a correlation between the objective and subjective evaluation exists in analyzing soft tissue color. A total of 39 patients with at least one implant-supported restoration in the anterior maxilla were included in the study. The shade of the peri-implant mucosa was compared with the shade of the gingiva at the adjacent tooth in a subjective and in an objective manner. The subjective evaluation was performed by five dental professionals (prosthodontist, periodontist, general dentist, dental hygienist, and dental assistant) in a subjective scale (ranging from 1 to 4). The objective evaluation was obtained by means of a spectrophotometer in a CIELAB* Color Scale, and the differences were evaluated through formula ΔE=[(ΔL)²+(Δa)²+(Δb)²]¹/². To correlate the subjective and the objective evaluation, for each arithmetical median value of the subjective evaluation, a mean value of objective evaluation has been calculated, and the Spearman's rank correlation coefficient has been used. The differences have been also analyzed for thin and thick tissue biotypes. The mean ∆E value for the subjective evaluation between peri-implant soft tissue and adjacent tooth gingival tissue was ∆E = 9.74. Also, mean ∆E values of 10.35 and 7.54 have been reported for thin and thick biotypes, respectively. Mean values of ∆E = 6.63, 8.54, and 15.54 were presented by median values of 1 (perfect matching), 2 (good matching), and 3 (clinically distinguishable), respectively. The threshold for the distinction of differences of mucosal color by the human eyes between perfect or good matching and distinguishable values has been calculated in ∆E = 8.74. Within the limitation of this study, a correlation between the subjective and the objective evaluation of the peri-implant soft tissue exists and the threshold for the distinction of mucosal color differences between perfect or good matching and distinguishable subjective values has been calculated in ∆E = 8.74 in the objective evaluation.
- Research Article
3
- 10.1061/jhtrcq.0000659
- Dec 1, 2018
- Journal of Highway and Transportation Research and Development (English Edition)
The relationship between subjective and objective evaluations on the steering performance of passenger vehicles is investigated to realize the subjective evaluation with objective measurement and to completely replace the objective measurement by subjective evaluation. The subjective evaluation uses steering performance as the study point, and objective evaluation focuses on angle pulse, wheel, snaking, and central tests. The subjective and objective evaluation systems are established, which consist of 7 subjective evaluation indicators (inherent steering characteristics, steering transient response, yawing response, and others) and 17 objective evaluation indicators (formant frequency, average steering angle, yaw velocity gain, et al.). A total of 7 groups of subjective evaluation score and 17 groups of objective evaluation score are obtained by evaluating 4 vehicles. The functional relationship between subjective and objective evaluations is established through regression analysis. Result shows that a 1D binary linear relationship exists between subjective and objective evaluation indicators under a small sample condition. Most of the subjective evaluation indicators have a negative relationship with formant level within the linear function. The formant level is a critical indicator that influences the steering performance of passenger vehicles on the basis of hypothesis testing and dynamic analysis. The lower the formant level is, the higher the subjective evaluation score will be.
- Research Article
29
- 10.3390/cancers12102771
- Sep 27, 2020
- Cancers
Simple SummaryOral cancer has a high mortality rate. Then, oral cancer screening is needed for early detection and treatment. Fluorescence visualization is non-invasive, convenient, and in real-time, and examinations can be repeated. Our study aimed to show the usefulness of oral cancer screening with fluorescence visualization. A total of 502 patients were performed using fluorescence visualization that was analyzed using subjective and objective evaluation. Results of this study, subjective evaluation for detection oral cancer was high sensitivity and low specificity, while objective evaluation using imaging processing analysis was high sensitivity and high specificity. Therefore, oral cancer screening using fluorescence visualization is useful for the detection of oral cancer. The widespread use of this screening can reduce the mortality rate of oral cancer.Background: Oral cancer screening is important for early detection and early treatment, which help improve survival rates. Biopsy is the gold standard for a definitive diagnosis but is invasive and painful, while fluorescence visualization is non-invasive, convenient, and real-time, and examinations can be repeated using optical instruments. The purpose of this study was to clarify the usefulness of fluorescence visualization in oral cancer screening. Methods: A total of 502 patients, who were examined using fluorescence visualization with optical instruments in our hospitals between 2014 and 2019, were enrolled in this study. The final diagnosis was performed by pathological examination. Fluorescence visualization was analyzed using subjective and objective evaluations. Results: Subjective evaluations for detecting oral cancer offered 96.8% sensitivity and 48.4% specificity. Regarding the objective evaluations, sensitivity and specificity were 43.7% and 84.6% for mean green value, 55.2% and 67.0% for median green value, 82.0% and 44.2% for coefficient of variation of value, 59.6% and 45.3% for skewness, and 85.1% and 75.8% for value ratio. For the sub-analysis of oral cancer, all factors on objective and subjective evaluation showed no significant difference. Conclusions: Fluorescence visualization with subjective and objective evaluation is useful for oral cancer screening.
- Research Article
- 10.33005/ic-ebgc.v1i1.11
- Aug 15, 2022
- Proceedings of International Conference on Economics Business and Government Challenges
Previous research found the use of post-linguistic features (emotionality, complexity, and informality) affect social media communicators success in persuading and attracting audiences. This study aims to explore the differences in linguistic styles used in persuasive communication adopted by capital market social media influencers. Using 5.710 instagram posts from a sample of five influencers and 4.149 instagram posts from a sample of five securities companies, we applied a text mining method to identify linguistic features used by the influencers and compare the characteristics of linguistic features used by individual influencers and institutional influencers. The results suggest that individual influencers adopt more persuasive linguistic model than institution influencers. This study contributes to the literature pertaining to social media influence strategy, particularly in describing the linguistic features used by social media influencers (individual and institutional influencers). Keywords: Communication; Linguistic style; Social media content; Text mining
- Conference Article
7
- 10.5167/uzh-28555
- May 1, 2000
- Zurich Open Repository and Archive (University of Zurich)
Evaluating translation quality as input to product development
- Research Article
22
- 10.1109/access.2020.2980075
- Jan 1, 2020
- IEEE Access
In order to enhance viewing experiences, a number of backlight local dimming (BLD) algorithms have been developed to improve the image contrast ratio and provide power efficiency for modern displays. In order to evaluate which BLD algorithm performs best for HDR images rendering on dual-panel displays, this paper presents a comprehensive subjective and objective evaluation conducted with five BLD algorithms across a number of scenes. The subjective evaluation (N = 24) required participants to rank each BLD generated image based on which they thought was the most natural looking. The objective evaluation was undertaken via the use of a novel methodology to generate the images per BLD for comparison against the ground truth High Dynamic Range (HDR) image. Resulting images were compared with the ground truth using qualitative methods namely: HDR-VDP, puPSNR, puSSIM and puVIFP. The power-saving rate of each method was also calculated. The results demonstrate a strong correlation between objective and subjective evaluation. Furthermore, results show that BLD algorithms that consider the luminance balance between backlight and LCD images perform better than straightforward BLD methods.
- Research Article
- 10.15421/462116
- Feb 22, 2022
- Journal “Ukrainian sense”
Abstract
 Background. One of the most important problems of modern Ukrainian linguistic studies is the comprehensive study of the language as a dynamic system, which has recently been noticeably «updated» at all levels. Under the influence of a number of social factors the lexical-semantic and derivational systems change most noticeably, the dynamics of other linguistic systems, morphological system in particular. In various socio-communicative fields, the functioning of many variabilities of grammatical forms of verbs that show the category of aspect of the verb is observed. New tendencies in the formation of perfectives and imperfectives in this regard are the most significant, they are most active in informational and publicistic, scientific, literary discourses.
 The purpose of the article is to identify and prove specific features of perfectivations / imperfectivation of aspectual pairs of verbs in the language of prose of P. Zahrebelnyi The main tasks of the article: to describe the writer's perfectives and imperfectives as stylisthemes that in the literary discourse of the prose writer represent the category of occasionality; specify the methods of their derivation; identify the main functions of stylisthemes in the author's novel. 
 Methods. In this publication were used general scientific methods such as observation, analysis, synthesis and comparison; within linguistic methods, there is the descriptive method which is used for interpretation of perfectivation / imperfectivation in the literary discourse of P. Zagrebelny; method of integrative analysis is the basis of the overall study of the set of stylistic matters that represent the grammatical category of aspect in the writer’s language; method of linguistic analysis allows to determine stylistic functions of author’s individual perfectives and imperfectives in the vertical context of analyzed novels. 
 Results. In the morphological system of the modern Ukrainian literary language, not all verbs form aspectual pairs. One-aspect (non-biaspectual) verbs form quite a steady subsystem within the verbal system of the Ukrainian language. However, within grammatical verb categories span different points of tense, aspect, state, the person of a verb, functioning of active and passive particles became visibly active from the end of the 80s – beginning of the 90s of XX century. At the beginning of the 20s of the XXI century we can see consequences of tendencies that not only show the dynamic nature of the verbal system but also the dynamic consistency of grammatical norms.
 In the language of P. Zagrebelny's fiction, potential verbs, identifying the category of occasionality, reflect modern tendencies in the development of the Ukrainian language and can strengthen the potential of the literary standard. In the works of the author, written in the 60’s – the early ‘80s of last century, grammatical forms are found, which represent the perfection of verbs of imperfect form and the imperfection of verbs of perfect form, which in modern Ukrainian became more active only in the late 80’s – early ‘90s of the twentieth century. A specific feature of the writer’s linguistic style is the active production of aspectual pairs, which in the vertical context of his language creative work acts as a stylistheme. The whole set of individual author's perfectives and imperfectives is a consequence of an unusual (previously untraceable) combination of usual morphemes.
 Discussion. Numerous potential verb forms that mark the literary discourse of P. Zahrebelnyi demonstrate the author's linguistic sense to choose the ways of development of the Ukrainian language, the knowledge of tendencies of activation of its grammatical phenomena during the last few decades in particular. From the '60s of the twentieth century, the writer actualized numerous forms within verb grammatical categories, in particular the verb category of the form, which mostly corresponds to the modern linguistic and literary standard, significantly diversified the language of fiction with stylisthemes that are potentially close to commonly used linguistic units. We see prospects for further research of this topical problem in a comprehensive study of individual author's formation within the verb categories of state and personality / impersonality, which also reflect the dynamics of the stylistic norm of the modern Ukrainian literary language, represented in literary discourse.
 Keywords: grammatical category of aspect, perfective, imperfective, stylistic norm, literary discourse.
- Research Article
17
- 10.1243/09544070jauto563
- Mar 1, 2008
- Proceedings of the Institution of Mechanical Engineers, Part D: Journal of Automobile Engineering
The gear whine sound of an axle system is one of the most important sound qualities in a sports utility vehicle (SUV). Previous work has shown that, because of masking effects, it is difficult to evaluate the gear whine sound objectively by using only the A-weighted sound pressure level. In this paper, the characteristics of the axle-gear whine sound were first investigated on the basis of synthetic sound technology, and a new objective evaluation method for this sound was developed by using sound metrics, which are the psychoacoustic parameters, and the artificial neural network (ANN) used for the modelling of the correlation between objective and subjective evaluation. This model developed by using ANN was applied to the objective evaluation of the axle-gear whine sound for real SUVs and the output of the model was compared with subjective evaluation. The results indicate a good correlation of over 90 per cent between the subjective and objective evaluations.