A novel code representation for detecting Java code clones using high-level and abstract compiled code representations.

In software development, it's common to reuse existing source code by copying and pasting, resulting in the proliferation of numerous code clones-similar or identical code fragments-that detrimentally affect software quality and maintainability. Although several techniques for code clone detection exist, many encounter challenges in effectively identifying semantic clones due to their inability to extract syntax and semantics information. Fewer techniques leverage low-level source code representations like bytecode or assembly for clone detection. This work introduces a novel code representation for identifying syntactic and semantic clones in Java source code. It integrates high-level features extracted from the Abstract Syntax Tree with low-level features derived from intermediate representations generated by static analysis tools, like the Soot framework. Leveraging this combined representation, fifteen machine-learning models are trained to effectively detect code clones. Evaluation on a large dataset demonstrates the models' efficacy in accurately identifying semantic clones. Among these classifiers, ensemble classifiers, such as the LightGBM classifier, exhibit exceptional accuracy. Linearly combining features enhances the effectiveness of the models compared to multiplication and distance combination techniques. The experimental findings indicate that the proposed method can outperform the current clone detection techniques in detecting semantic clones.

Animal cognition: Dogs build semantic expectations between spoken words and objects.

A recent study has used scalp-recorded electroencephalography to obtain evidence of semantic processing of human speech and objects by domesticated dogs. The results suggest that dogs do comprehend the meaning of familiar spoken words, in that a word can evoke the mental representation of the object to which it refers.

Universal and cultural factors shape body part vocabularies.

Every human has a body. Yet, languages differ in how they divide the body into parts to name them. While universal naming strategies exist, there is also variation in the vocabularies of body parts across languages. In this study, we investigate the similarities and differences in naming two separate body parts with one word, i.e., colexifications. We use a computational approach to create networks of body part vocabularies across languages. The analyses focus on body part networks in large language families, on perceptual features that lead to colexifications of body parts, and on a comparison of network structures in different semantic domains. Our results show that adjacent body parts are colexified frequently. However, preferences for perceptual features such as shape and function lead to variations in body part vocabularies. In addition, body part colexification networks are less varied across language families than networks in the semantic domains of emotion and colour. The study presents the first large-scale comparison of body part vocabularies in 1,028 language varieties and provides important insights into the variability of a universal human domain.

Elucidating the semantics-topology trade-off for knowledge inference-based pharmacological discovery.

Leveraging AI for synthesizing the deluge of biomedical knowledge has great potential for pharmacological discovery with applications including developing new therapeutics for untreated diseases and repurposing drugs as emergent pandemic treatments. Creating knowledge graph representations of interacting drugs, diseases, genes, and proteins enables discovery via embedding-based ML approaches and link prediction. Previously, it has been shown that these predictive methods are susceptible to biases from network structure, namely that they are driven not by discovering nuanced biological understanding of mechanisms, but based on high-degree hub nodes. In this work, we study the confounding effect of network topology on biological relation semantics by creating an experimental pipeline of knowledge graph semantic and topological perturbations. We show that the drop in drug repurposing performance from ablating meaningful semantics increases by 21% and 38% when mitigating topological bias in two networks. We demonstrate that new methods for representing knowledge and inferring new knowledge must be developed for making use of biomedical semantics for pharmacological innovation, and we suggest fruitful avenues for their development.

Learning the Meanings of Function Words From Grounded Language Using a Visual Question Answering Model.

Interpreting a seemingly simple function word like "or," "behind," or "more" can require logical, numerical, and relational reasoning. How are such words learned by children? Prior acquisition theories have often relied on positing a foundation of innate knowledge. Yet recent neural-network-based visual question answering models apparently can learn to use function words as part of answering questions about complex visual scenes. In this paper, we study what these models learn about function words, in the hope of better understanding how the meanings of these words can be learned by both models and children. We show that recurrent models trained on visually grounded language learn gradient semantics for function words requiring spatial and numerical reasoning. Furthermore, we find that these models can learn the meanings of logical connectives and and or without any prior knowledge of logical reasoning as well as early evidence that they are sensitive to alternative expressions when interpreting language. Finally, we show that word learning difficulty is dependent on the frequency of models' input. Our findings offer proof-of-concept evidence that it is possible to learn the nuanced interpretations of function words in a visually grounded context by using non-symbolic general statistical learning algorithms, without any prior knowledge of linguistic meaning.

Wh-Words: Existential or Universal Quantifiers in Child Mandarin?

Wh-words have been analysed as existential quantifiers (Chierchia in Logic in grammar: polarity, free choice, and intervention. Oxford University Press, Oxford, 2013; Fox, in Sauerland U, Stateva P (eds) Presupposition and implicature in compositional semantics (Palgrave studies in pragmatics, language and cognition). Palgrave MacMillan, Houndmills, pp 71-120, 2007; Liao in Alternative and exhaustification: non-interrogative uses of Chinese wh-words. Harvard University, 2010) or universal quantifiers (Nishigauchi, in: Theoretical and applied linguistics at Kobe Shoin 2, Kobe Shoin Institute for Linguistic Sciences, 1999). These two accounts have distinct predictions on how children initially interpret wh-words. The universal account predicts that children should initially interpret wh-words as universal quantifiers, whereas the existential account anticipates that children should start out with the existential interpretation. To adjudicate between the two accounts, the present study was designed to explore pre-schoolers' semantic knowledge of wh-quantification. Specifically, it investigated the interpretation of the wh-word shenme 'what' with 4-and 5-year-old Mandarin-speaking children and a control group of adults. Using a Truth Value Judgment Task (Crain and Thornton in Investigations in universal grammar: a guide to experiments on the acquisition of syntax and semantics. MIT Press, Cambridge, 1998), Experiment 1 evaluated whether children interpret the wh-word shenme 'what' as closer in meaning to the polarity sensitive item renhe 'any' or the universal quantifier suoyou 'all' in the antecedent of ruguo 'if' conditionals. Using a Question-Answer Task, Experiments 2 & 3 respectively investigated whether children interpret shenme 'what' as closer in meaning to renhe 'any' or suoyou 'all' in two types of questions: yes-no questions with the particle ma and A-not-A questions. It was found that both children and adults interpret shenme 'what' as closer in meaning to renhe 'any' than suoyou 'all'. The findings suggest that Mandarin-speaking pre-schoolers already have adult-like semantic knowledge of wh-quantification: wh-words are existential quantifiers rather than universal quantifiers. Due to the paucity of primary linguistic input, children's early mastery of the non-interrogative wh-words appear to support the biolinguistic approach to language acquisition (Chomsky in Aspects of the theory of syntax. MIT Press, Cambridge, 1965; Pinker in Language learnability and language development. Harvard University Press, Cambridge, 1984; Crain et al. in Language acquisition from a biolinguistic perspective. Neurosci Biobehav Rev, 2016. https://doi.org/10.1016/j.neubiorev.2016.09.004 ).

It Costs To Exist: Acceptability Judgments of the Temporal Concord of the Auxiliary Verbs You and Hui in Taiwan Mandarin.

The mechanisms underlying the processing of the temporal reference of a sentence are still unexplored. Most of the previous psycholinguistic studies used the temporal concord violation between deictic time adverbs and tense marking on the verb to investigate this issue. They found that processing past tense marking is more difficult than non-past tense, indicated by lower accuracy rates and/or longer reaction time. However, it is not clear whether this complexity is due to tense marking or the temporal reference it denotes. This paper examines this issue with a judgment acceptability experiment in Taiwan Mandarin, which is analyzed as a tenseless language. The two modal auxiliary verbs you and hui were placed after deictic past time adverbs (grammatical with you but not with hui) and deictic future time adverbs (grammatical with hui but not with you). The temporal concord violation of the auxiliary verb you led to higher acceptability rates but longer reaction time than hui, reflecting higher processing difficulties. This paper argues that these complexities are due to the existential-assertive meaning of you, which interplays with the meaning of the event described by the verb rendering the situation more or less likely to occur in the future. The computation of the temporal concord of hui, displaying a future sense meaning, is more straightforward and therefore easier to process. This suggests that the mechanisms responsible for temporal reference processing are of different nature depending on the semantics of the temporal marker in the sentence.

Asynchronous behavioral and neurophysiological changes in word production in the adult lifespan.

Behavioral and brain-related changes in word production have been claimed to predominantly occur after 70 years of age. Most studies investigating age-related changes in adulthood only compared young to older adults, failing to determine whether neural processes underlying word production change at an earlier age than observed in behavior. This study aims to fill this gap by investigating whether changes in neurophysiological processes underlying word production are aligned with behavioral changes. Behavior and the electrophysiological event-related potential patterns of word production were assessed during a picture naming task in 95 participants across five adult lifespan age groups (ranging from 16 to 80 years old). While behavioral performance decreased starting from 70 years of age, significant neurophysiological changes were present at the age of 40 years old, in a time window (between 150 and 220 ms) likely associated with lexical-semantic processes underlying referential word production. These results show that neurophysiological modifications precede the behavioral changes in language production; they can be interpreted in line with the suggestion that the lexical-semantic reorganization in mid-adulthood influences the maintenance of language skills longer than for other cognitive functions.

Default mode network shows distinct emotional and contextual responses yet common effects of retrieval demands across tasks.

The default mode network (DMN) lies towards the heteromodal end of the principal gradient of intrinsic connectivity, maximally separated from the sensory-motor cortex. It supports memory-based cognition, including the capacity to retrieve conceptual and evaluative information from sensory inputs, and to generate meaningful states internally; however, the functional organisation of DMN that can support these distinct modes of retrieval remains unclear. We used fMRI to examine whether activation within subsystems of DMN differed as a function of retrieval demands, or the type of association to be retrieved, or both. In a picture association task, participants retrieved semantic associations that were either contextual or emotional in nature. Participants were asked to avoid generating episodic associations. In the generate phase, these associations were retrieved from a novel picture, while in the switch phase, participants retrieved a new association for the same image. Semantic context and emotion trials were associated with dissociable DMN subnetworks, indicating that a key dimension of DMN organisation relates to the type of association being accessed. The frontotemporal and medial temporal DMN showed a preference for emotional and semantic contextual associations, respectively. Relative to the generate phase, the switch phase recruited clusters closer to the heteromodal apex of the principal gradient-a cortical hierarchy separating unimodal and heteromodal regions. There were no differences in this effect between association types. Instead, memory switching was associated with a distinct subnetwork associated with controlled internal cognition. These findings delineate distinct patterns of DMN recruitment for different kinds of associations yet common responses across tasks that reflect retrieval demands.

UMobileNetV2 model for semantic segmentation of gastrointestinal tract in MRI scans.

Gastrointestinal (GI) cancer is leading general tumour in the Gastrointestinal tract, which is fourth significant reason of tumour death in men and women. The common cure for GI cancer is radiation treatment, which contains directing a high-energy X-ray beam onto the tumor while avoiding healthy organs. To provide high dosages of X-rays, a system needs for accurately segmenting the GI tract organs. The study presents a UMobileNetV2 model for semantic segmentation of small and large intestine and stomach in MRI images of the GI tract. The model uses MobileNetV2 as an encoder in the contraction path and UNet layers as a decoder in the expansion path. The UW-Madison database, which contains MRI scans from 85 patients and 38,496 images, is used for evaluation. This automated technology has the capability to enhance the pace of cancer therapy by aiding the radio oncologist in the process of segmenting the organs of the GI tract. The UMobileNetV2 model is compared to three transfer learning models: Xception, ResNet 101, and NASNet mobile, which are used as encoders in UNet architecture. The model is analyzed using three distinct optimizers, i.e., Adam, RMS, and SGD. The UMobileNetV2 model with the combination of Adam optimizer outperforms all other transfer learning models. It obtains a dice coefficient of 0.8984, an IoU of 0.8697, and a validation loss of 0.1310, proving its ability to reliably segment the stomach and intestines in MRI images of gastrointestinal cancer patients.

Assessment of changes in vessel area during needle manipulation in microvascular anastomosis using a deep learning-based semantic segmentation algorithm: A pilot study.

Appropriate needle manipulation to avoid abrupt deformation of fragile vessels is a critical determinant of the success of microvascular anastomosis. However, no study has yet evaluated the area changes in surgical objects using surgical videos. The present study therefore aimed to develop a deep learning-based semantic segmentation algorithm to assess the area change of vessels during microvascular anastomosis for objective surgical skill assessment with regard to the "respect for tissue." The semantic segmentation algorithm was trained based on a ResNet-50 network using microvascular end-to-side anastomosis training videos with artificial blood vessels. Using the created model, video parameters during a single stitch completion task, including the coefficient of variation of vessel area (CV-VA), relative change in vessel area per unit time (ΔVA), and the number of tissue deformation errors (TDE), as defined by a ΔVA threshold, were compared between expert and novice surgeons. A high validation accuracy (99.1%) and Intersection over Union (0.93) were obtained for the auto-segmentation model. During the single-stitch task, the expert surgeons displayed lower values of CV-VA (p < 0.05) and ΔVA (p < 0.05). Additionally, experts committed significantly fewer TDEs than novices (p < 0.05), and completed the task in a shorter time (p < 0.01). Receiver operating curve analyses indicated relatively strong discriminative capabilities for each video parameter and task completion time, while the combined use of the task completion time and video parameters demonstrated complete discriminative power between experts and novices. In conclusion, the assessment of changes in the vessel area during microvascular anastomosis using a deep learning-based semantic segmentation algorithm is presented as a novel concept for evaluating microsurgical performance. This will be useful in future computer-aided devices to enhance surgical education and patient safety.

Joint extraction of wheat germplasm information entity relationship based on deep character and word fusion.

The verified text data of wheat varieties is an important component of wheat germplasm information. To automatically obtain a structured description of the phenotypic and genetic characteristics of wheat varieties, the aim at solve the issues of fuzzy entity boundaries and overlapping relationships in unstructured wheat variety approval data, WGIE-DCWF (joint extraction model of wheat germplasm information entity relationship based on deep character and word fusion) was proposed. The encoding layer of the model deeply fused word semantic information and character information using the Transformer encoder of BERT. This allowed for the cascading fusion of contextual semantic feature information to achieve rich character vector representation and improve the recognition ability of entity features. The triple extraction layer of the model established a cascading pointer network, extracted the head entity, extracted the tail entity according to the relationship category, and decoded the output triplet. This approach improved the model's capability to extract overlapping relationships. The experimental results demonstrated that the WGIE-DCWF model performed exceptionally well on both the WGD (wheat germplasm dataset) and the public dataset DuIE. The WGIE-DCWF model not only achieved high performance on the evaluation datasets but also demonstrated good generalization. This provided valuable technical support for the construction of a wheat germplasm information knowledge base and is of great significance for wheat breeding, genetic research, cultivation management, and agricultural production.

Differential neural correlates underlying visuospatial versus semantic reasoning in autistic children.

While fronto-posterior underconnectivity has often been reported in autism, it was shown that different contexts may modulate between-group differences in functional connectivity. Here, we assessed how different task paradigms modulate functional connectivity differences in a young autistic sample relative to typically developing children. Twenty-three autistic and 23 typically developing children aged 6 to 15 years underwent functional magnetic resonance imaging (fMRI) scanning while completing a reasoning task with visuospatial versus semantic content. We observed distinct connectivity patterns in autistic versus typical children as a function of task type (visuospatial vs. semantic) and problem complexity (visual matching vs. reasoning), despite similar performance. For semantic reasoning problems, there was no significant between-group differences in connectivity. However, during visuospatial reasoning problems, we observed occipital-occipital, occipital-temporal, and occipital-frontal over-connectivity in autistic children relative to typical children. Also, increasing the complexity of visuospatial problems resulted in increased functional connectivity between occipital, posterior (temporal), and anterior (frontal) brain regions in autistic participants, more so than in typical children. Our results add to several studies now demonstrating that the connectivity alterations in autistic relative to neurotypical individuals are much more complex than previously thought and depend on both task type and task complexity and their respective underlying cognitive processes.

Evaluation of adjective and adverb types for effective Twitter sentiment classification.

Twitter, the largest microblogging platform, has reported more than 330 million active users in recent years. Many users express their sentiments about politics, sports, products, personalities, etc. Sentiment analysis has emerged as a specialized branch of machine learning in which tweets are binary-classified to provide sentimental insights. A major step in sentiment classification is feature selection, which primarily revolves around parts of speech (POS). Few techniques merely focused on single features such as adjectives, adverbs, and verbs, while other techniques examined types of these features, such as comparative adjectives, superlative adjectives, or general adverbs. Furthermore, POS as linguistic entities have also been studied and extensively classified by researchers, such as CLAWS-C7. For sentiment analysis, none of the studies conceptualized all possible POS features under similar conditions to draw firm conclusion. This research is centered on the following objectives: 1) examining the impact of various types of adjectives and adverbs that have not been previously explored for sentiment classification; 2) analyzing potential combinations of adjectives and adverbs types 3) conducting a comparison with a benchmark dataset for better classification accuracy. To assess the concept, a renowned human annotated dataset of tweets is investigated. Results showed that classification accuracy for adjectives is improved up to 83% based on the general superlative adjective whereas for adverbs, comparative general adverb also depicted significant accuracy improvement. Their combination with general adjectives and general adverbs also played a substantial role. The unexplored potential of adjectives and adverb types proved better in accuracy against state-of-the-art probabilistic model. In comparison to lexicon-based model, proposed research model overruled the dependency of lexicon-based dictionary where each term first needs to be matched for semantic orientation. The evident outcomes also help in time reduction aspect where huge volume of data need to be processed swiftly. This noteworthy contribution brought up significant knowledge and direction for domain experts. In the future, the proposed technique will be explored for other types of textual data across different domains.

Distinguishing the Uterine Artery, the Ureter, and Nerves in Laparoscopic Surgical Images Using Ensembles of Binary Semantic Segmentation Networks.

Performing a minimally invasive surgery comes with a significant advantage regarding rehabilitating the patient after the operation. But it also causes difficulties, mainly for the surgeon or expert who performs the surgical intervention, since only visual information is available and they cannot use their tactile senses during keyhole surgeries. This is the case with laparoscopic hysterectomy since some organs are also difficult to distinguish based on visual information, making laparoscope-based hysterectomy challenging. In this paper, we propose a solution based on semantic segmentation, which can create pixel-accurate predictions of surgical images and differentiate the uterine arteries, ureters, and nerves. We trained three binary semantic segmentation models based on the U-Net architecture with the EfficientNet-b3 encoder; then, we developed two ensemble techniques that enhanced the segmentation performance. Our pixel-wise ensemble examines the segmentation map of the binary networks on the lowest level of pixels. The other algorithm developed is a region-based ensemble technique that takes this examination to a higher level and makes the ensemble based on every connected component detected by the binary segmentation networks. We also introduced and trained a classic multi-class semantic segmentation model as a reference and compared it to the ensemble-based approaches. We used 586 manually annotated images from 38 surgical videos for this research and published this dataset.

A Dataset for Evaluating Contextualized Representation of Biomedical Concepts in Language Models.

Due to the complexity of the biomedical domain, the ability to capture semantically meaningful representations of terms in context is a long-standing challenge. Despite important progress in the past years, no evaluation benchmark has been developed to evaluate how well language models represent biomedical concepts according to their corresponding context. Inspired by the Word-in-Context (WiC) benchmark, in which word sense disambiguation is reformulated as a binary classification task, we propose a novel dataset, BioWiC, to evaluate the ability of language models to encode biomedical terms in context. BioWiC comprises 20'156 instances, covering over 7'400 unique biomedical terms, making it the largest WiC dataset in the biomedical domain. We evaluate BioWiC both intrinsically and extrinsically and show that it could be used as a reliable benchmark for evaluating context-dependent embeddings in biomedical corpora. In addition, we conduct several experiments using a variety of discriminative and generative large language models to establish robust baselines that can serve as a foundation for future research.

Irregular word reading as a marker of semantic decline in Alzheimer's disease: implications for premorbid intellectual ability measurement.

BACKGROUND: Irregular word reading has been used to estimate premorbid intelligence in Alzheimer's disease (AD) dementia. However, reading models highlight the core influence of semantic abilities on irregular word reading, which shows early decline in AD. The primary objective of this study is to ascertain whether irregular word reading serves as an indicator of cognitive and semantic decline in AD, potentially discouraging its use as a marker for premorbid intellectual abilities. METHOD: Six hundred eighty-one healthy controls (HC), 104 subjective cognitive decline, 290 early and 589 late mild cognitive impairment (EMCI, LMCI) and 348 AD participants from the Alzheimer's Disease Neuroimaging Initiative were included. Irregular word reading was assessed with the American National Adult Reading Test (AmNART). Multiple linear regressions were conducted predicting AmNART score using diagnostic category, general cognitive impairment and semantic tests. A generalized logistic mixed-effects model predicted correct reading using extracted psycholinguistic characteristics of each AmNART words. Deformation-based morphometry was used to assess the relationship between AmNART scores and voxel-wise brain volumes, as well as with the volume of a region of interest placed in the left anterior temporal lobe (ATL), a region implicated in semantic memory. RESULTS: EMCI, LMCI and AD patients made significantly more errors in reading irregular words compared to HC, and AD patients made more errors than all other groups. Across the AD continuum, as well as within each diagnostic group, irregular word reading was significantly correlated to measures of general cognitive impairment / dementia severity. Neuropsychological tests of lexicosemantics were moderately correlated to irregular word reading whilst executive functioning and episodic memory were respectively weakly and not correlated. Age of acquisition, a primarily semantic variable, had a strong effect on irregular word reading accuracy whilst none of the phonological variables significantly contributed. Neuroimaging analyses pointed to bilateral hippocampal and left ATL volume loss as the main contributors to decreased irregular word reading performances. CONCLUSIONS: While the AmNART may be appropriate to measure premorbid intellectual abilities in cognitively unimpaired individuals, our results suggest that it captures current semantic decline in MCI and AD patients and may therefore underestimate premorbid intelligence. On the other hand, irregular word reading tests might be clinically useful to detect semantic impairments in individuals on the AD continuum.

The Effects of Problematic Internet Use and Emotional Connotation on Internet Slang Processing: Evidence from a Lexical Decision Task.

Young people use slang for identifying themselves with a particular social group, gaining social recognition and respect from that group, and expressing their emotional state. One feature of Internet slang is its active use by youth in online communication, which, under certain conditions, may cause problematic Internet use (PIU). We conducted two studies in young Russian speakers (n1 = 115, n2 = 106). In study 1, participants were asked to rate a set of slang and common words using Self-Assessment Manikin. The study revealed that the most reliable predictor of higher emotional ratings was word familiarity. There were no significant effects of slang vs. common words or word frequency. In study 2, we used a dual lexical decision task to reveal the effects of word characteristics and propensity for PIU on reaction time (RT) for Internet slang words in pairs with semantically related vs. unrelated common words. Study 2 did not reveal any significant semantic priming effect. Word frequency was a significant predictor of lexical decision facilitation. Common, but not slang, word valence and dominance significantly affected RT in the opposite direction. Individuals with higher cognitive preoccupation with the Internet responded significantly faster, while those more likely to use online communication for mood regulation responded significantly slower to the stimuli. Apparently, on explicit and implicit levels, in-depth knowledge of Internet slang can be one the PIU markers. The results are discussed in line with Davis' approach to determining the general pathological Internet use.

KDGene: knowledge graph completion for disease gene prediction using interactional tensor decomposition.

The accurate identification of disease-associated genes is crucial for understanding the molecular mechanisms underlying various diseases. Most current methods focus on constructing biological networks and utilizing machine learning, particularly deep learning, to identify disease genes. However, these methods overlook complex relations among entities in biological knowledge graphs. Such information has been successfully applied in other areas of life science research, demonstrating their effectiveness. Knowledge graph embedding methods can learn the semantic information of different relations within the knowledge graphs. Nonetheless, the performance of existing representation learning techniques, when applied to domain-specific biological data, remains suboptimal. To solve these problems, we construct a biological knowledge graph centered on diseases and genes, and develop an end-to-end knowledge graph completion framework for disease gene prediction using interactional tensor decomposition named KDGene. KDGene incorporates an interaction module that bridges entity and relation embeddings within tensor decomposition, aiming to improve the representation of semantically similar concepts in specific domains and enhance the ability to accurately predict disease genes. Experimental results show that KDGene significantly outperforms state-of-the-art algorithms, whether existing disease gene prediction methods or knowledge graph embedding methods for general domains. Moreover, the comprehensive biological analysis of the predicted results further validates KDGene's capability to accurately identify new candidate genes. This work proposes a scalable knowledge graph completion framework to identify disease candidate genes, from which the results are promising to provide valuable references for further wet experiments. Data and source codes are available at https://github.com/2020MEAI/KDGene.

[Optimization of prediction model for personalized water pills based on semantic analysis of traditional Chinese medicine materials].

This study aims to optimize the prediction model of personalized water pills that has been established by our research group. Dioscoreae Rhizoma, Leonuri Herba, Codonopsis Radix, Armeniacae Semen Amarum, and calcined Oyster were selected as model medicines of powdery, fibrous, sugary, oily, and brittle materials, respectively. The model prescriptions were obtained by uniform mixing design. With hydroxypropyl methylcellulose E5(HPMC-E5) aqueous solution as the adhesive, personalized water pills were prepared by extrusion and spheronizaition. The evaluation indexes in the pill preparation process and the multi-model statistical analysis were employed to optimize and evaluate the prediction model of personalized water pills. The prediction equation of the adhesive concentration was obtained as follows: Y_1=-4.172+3.63X_A+15.057X_B+1.838X_C-0.997X_D(adhesive concentration of 10% when Y_1<0, and 20% when Y_1>0). The overall accuracy of the prediction model for adhesive concentration was 96.0%. The prediction equation of adhesive dosage was Y_2=6.051+94.944X_A~(1.5)+161.977X_B+70.078X_C~2+12.016X_D~(0.3)+27.493X_E~(0.3)-2.168X_F~(-1)(R~2=0.954, P<0.001). Furthermore, the semantic prediction model for material classification of traditional Chinese medicines was used to classify the materials contained in the prescription, and thus the prediction model of personalized water pills was evaluated. The results showed that the prescriptions for model evaluation can be prepared with one-time molding, and the forming quality was better than that established by the research group earlier. This study has achieved the optimization of the prediction model of personalized water pills.