基于TF-IDF算法的运营商客户投诉原因研究

针对运营商人工处理客户投诉工单高成本低效率问题,提出了一种基于TF-IDF算法的定量研究方法,旨在高效精准地识别客户投诉原因。选用Jieba分词,导入自定义词典和停用词列表,对运营商客户投诉工单进行关键词抽取,获取各类问题中TF-IDF值排名前6的关键词,输出关键词集。提高了关键词抽取的准确性和效率。此外,对比仅对文档集使用TF进行统计和使用TextRank算法的情况,突显了IDF的重要性及算法原理的差异。实验结果表明,光猫、路由器、机顶盒问题广泛存在于各类投诉中。针对这三类问题,为运营商提供了改进产品、服务的相关建议,对运营商集中治理、解决问题具有一定的实用价值。

长三角一体化发展特征与动力探究——基于TF-IDF算法与格兰杰检验

从政策观念视角出发,深入探讨长三角区域一体化发展进程,有助于洞察区域内生发展动力和经济增长潜力。应用话语制度主义和间断-均衡框架,结合TF-IDF(词频-逆文档频率)算法与格兰杰检验,揭示长三角一体化发展特征与动因。研究发现:建设主体对5个发展目标的注意力分配不均衡;在创新共建目标方面,地方主体的注意力变动会引起中央主体的注意力变动;长三角一体化发展呈现小间断大均衡特征,体现“自下而上”的地方主导模式。

基于TF-IDF和多头注意力Transformer模型的文本情感分析

文本情感分析旨在对带有情感色彩的主观性文本进行分析、处理、归纳和推理,是自然语言处理中一项重要任务。针对现有的计算方法不能充分处理复杂度和混淆度较高的文本数据集的问题,提出了一种基于TF-IDF(Term Frequency-Inverse Document Frequency)和多头注意力Transformer模型的文本情感分析模型。在文本预处理阶段,利用TF-IDF算法对影响文本情感倾向较大的词语进行初步筛选,舍去常见的停用词及其他文本所属邻域对文本情感倾向影响较小的专有名词。然后,利用多头注意力Transformer模型编码器进行特征提取,抓取文本内部重要的语义信息,提高模型对语义的分析和泛化能力。该模型在多领域、多类型评论语料库数据集上取得了98.17%的准确率。

基于改进TF-IDF与BERT的领域情感词典构建方法

领域情感词典的构建是领域文本情感分析的基础。现有的领域情感词典构建方法存在所筛选候选情感词冗余度高、情感极性判断失准、领域依赖性强等问题。为了提高所筛选候选情感词的领域性和判断领域情感词极性的准确程度,提出了一种基于改进词频-逆文档频率(TF-IDF)与BERT的领域情感词典构建方法。该方法在筛选领域候选情感词阶段对TF-IDF算法进行改进,将隐含狄利克雷分布(LDA)算法与改进后的TF-IDF算法结合,进行领域性修正,提升了所筛选候选情感词的领域性;在候选情感词极性判断阶段,将情感倾向点互信息算法(SO-PMI)与BERT结合,利用领域情感词微调BERT分类模型,提高了判断领域候选情感词情感极性的准确程度。在不同领域的用户评论数据集上进行实验,结果表明,该方法可以提高所构建领域情感词典的质量,使用该方法构建的领域情感词典用于汽车领域和手机领域文本情感分析的F1值分别达到78.02%和88.35%。

Profile of Full-Term Births in Maternity Wards of Public Hospitals in Douala Cameroon

Introduction: Pregnancy as much as childbirth constitutes a risky situation, potentially fraught with sometimes dramatic complications: maternal death. Objective: We conducted this study with the aim of establishing the profile of those giving birth in our context with the aim to anticipate operationally in the future on morbidity but more on maternal deaths. Methodology: We conducted, using a structured questionnaire, a prospective descriptive study in representative maternity wards in the city of Douala;the study variables were socio-economic, anthropometric, obstetrical and clinical. Statistical analyses were carried out with CS Pro 7.3 and SPSS version 25.0 software. The Student, Chi-square and Fischer tests were used to compare the means of the variables and the percentages. Results: We recruited 305 births for our study. The average age of our births was 28.7 years ± 6.1 with an average height of 161.6 cm ± 5.06;an average body mass index at the start of pregnancy of 28.0 kilograms/square meter and 31.3 kilograms/square meter at delivery;the average weight gain was 8.4 g ± 5.37;an average gestation of 2.84±1.90;an average parity of 2.2 ± 2.1 with an average birth interval of 27.7 months ± 23.7. The average gestational age was 39.2 weeks ± 1.21 with pregnancy pathology dominated by malaria;85.9% began their prenatal follow-up before the 14th week of amenorrhea. Conclusion: The profile of childbirth in urban Cameroon does not seem potentially dystocic compared to that of the same regional and racial area.

A Comparison between Late Preterm and Term Infants with Respiratory Distress Syndrome, Early-Onset Sepsis, and Neonatal Jaundice in Ecuadorian Newborns

Background: To examine the differences in prevalence of respiratory distress syndrome, early-onset sepsis and jaundice, between late preterm infants versus term infants in Ecuadorian newborns. Methods: Study design: Epidemiological, observational, and cross-sectional, with two cohorts of patients. Settings: IESS Quito Sur Hospital at Quito, Ecuador, from February to April of 2020. Participants: This study included 204 newborns, 102 preterm infants, 102 term infants. Results: There are significant differences between late preterm infants and term infants, with a p-value of 0.000 in the prevalence of early sepsis, 70.59% vs. 35.29%. In respiratory distress syndrome between late and term premature infants, significant differences were observed with a p-value of 0.000, the proportion being 55.58% vs. 24.51% respectively. The prevalence of jaundice is higher in term infants with a p value of 0.002, 72.55%, versus 51.96% in late preterm infants, and the mean value of bilirubins in mg/dL was higher in term infants 14.32 versus 12.33 in late preterm infants;this difference is statistically significant with a p value of 0.004. Admission to the NICU is more frequent in late preterm infants with a p-value of 0.000, being 42.16% for late preterm infants vs. 7.84% in term infants;the mean of the hospital days with p-value 0.005, was higher in late preterm infants 4.97 days vs. 3.55 days for term newborns. Conclusion: Due to the conditions of their immaturity, late preterm infants are 2.86 times more likely to present early sepsis than full-term newborns. It is shown that late preterm infants are 2.69 times more likely to have respiratory distress syndrome compared to term infants, therefore, late preterm infants have a longer hospital stay of 4.97 days versus 3.55 days in term infants. Jaundice and mean bilirubin levels are higher in term infants due to blood group incompatibility and insufficient breastfeeding.

Long-term assessment of collagenase treatment for Dupuytren’s contracture:A 10-year follow-up study

BACKGROUND Enzymatic fasciotomy with collagenase clostridium histolyticum(CCH)has revolutionized the treatment for Dupuytren’s contracture(DC).Despite its benefits,the long-term outcomes remain unclear.This study presented a comprehensive 10-year follow-up assessment of the enduring effects of CCH on patients with DC.AIM To compare the short-term(12 wk)and long-term(10 years)outcomes on CCH treatment in patients with DC.METHODS A cohort of 45 patients was treated with CCH at the metacarpophalangeal(MCP)joint and the proximal interphalangeal(PIP)joint and underwent systematic reevaluation.The study adhered to multicenter trial protocols,and assessments were conducted at 12 wk,7 years,and 10 years post-surgery.RESULTS Thirty-seven patients completed the 10-year follow-up.At 10 years,patients treated at the PIP joint exhibited a 100%recurrence.However,patients treated at the MCP joint only showed a 50%recurrence.Patient satisfaction varied,with a lower satisfaction reported in PIP joint cases.Recurrence exceeding 20 degrees on the total passive extension deficit was observed,indicating a challenge for sustained efficacy.Significant differences were noted between outcomes at the 7-year and 10-year intervals.CONCLUSION CCH demonstrated sustained efficacy when applied to the MCP joint.However,caution is warranted for CCH treatment at the PIP joint due to a high level of recurrence and low patient satisfaction.Re-intervention is needed within a decade of treatment.

Long-term outcomes of endoscopic submucosal dissection for undifferentiated type early gastric cancer over 2 cm with R0 resection

BACKGROUND Endoscopic submucosal dissection(ESD)for over 2 cm in size undifferentiated type(UD type)early gastric cancer(EGC)confined to the mucosa is not only challenging,but also long-term outcomes are not well known.AIM To evaluate the long-term outcomes of ESD done for UD type EGCs confined to the mucosa over 2 cm in size and compare the results with those where the lesions were less than 2 cm.METHODS 143 patients with UD type EGC confirmed on histology after ESD at a tertiary hospital were reviewed.Cases with synchronous and metachronous lesions and a case with emergency surgery after ESD were excluded.A total of 137 cases were enrolled.79 cases who underwent R0 resection were divided into 2 cm or less(group A)and over 2 cm(group B)in size.RESULTS Among 79 patients who underwent R0 resection,the number in group A and B were 51 and 28,respectively.The mean follow-up period(SD)was 79.71±45.42 months.There was a local recurrence in group A(1/51,2%)and group B(1/28,3.6%)respectively.This patient in group A underwent surgery while the patient in group B underwent repeated ESD with no further recurrences in both patients.There was no regional lymph node metastasis,distant metastasis,and deaths in both groups.With R0 resection strategy for ESD on lesions over 2 cm,20.4%(28/137)of patients were able to avoid surgery compared with expanded indication.CONCLUSION If R0 resection is achieved by ESD,UD type EGCs over 2 cm also showed good and similar clinical outcomes as compared to lesions less than 2 cm when followed for over 5 years.With R0 resection strategy,several patients can avoid surgery.

一种融合改进TF-IDF与词典模型的情感分类算法

针对传统情感文本分类算法存在情感特征词的极性偏好区分度较低和稳定性较差等问题,提出了一种改进词频-逆文本频率(TF-IDF)模型与词典模型相融合的情感文本分类算法。首先,通过情感特征词在不同情感类型语料中的频率分布和离散系数,度量情感特征词极性偏好所包含的区分度和稳定性,生成情感特征词极性指标;然后,使用该指标改进TF-IDF模型的情感特征词权重;最后,基于改进的TF-IDF模型,使用带决策函数的有监督分类算法计算情感文本的极性得分,并与词典模型所得的极性得分进行调和平均,得到情感文本综合极性得分。

Draft of an Anthropometric Reference System for Full-Term Cameroonian Newborns: Prospective Study with Analytical Aim in the Maternity Wards of Douala

Introduction: Anthropometry applied to newborns is a reliable indicator of the quality of fetal growth. The latter is influenced by genetic, racial and nutritional factors varying from one population to another, explaining why a standard cannot be applied to all populations. Research question: should the Caucasian frame of reference be dogmatically applied in our African context? Multicenter studies are therefore necessary;hence the interest of this work, the main objective of which was to describe the anthropometric profile of full-term newborns in the city of Douala. Methodology: We carried out a cross-sectional study with an analytical aim and prospective data collection in the maternity wards of the Douala General Hospital, Laquintinie Hospital, District hospitals of Deido, Nylon and Bonassama over a period of 4 months (January to April 2020). We were interested in any newborn, born alive, vaginally or by cesarean section, seen in the first 24 hours from a full-term single-fetal pregnancy whose mother had given consent. We excluded newborns whose term was unclear and those with congenital malformations or signs of embryo-foetopathy. Data collection was done using structured and pre-tested survey sheets. The study variables were obstetric and anthropometric. Statistical analyzes were carried out with CS Pro 7.3 and SPSS version 25.0 software. The Student, Chi-square and Fischer tests were used to compare the means of the variables, the percentages with a significance threshold P value Results: During the study period, 305 full-term newborns were included, divided into 172 boys and 133 girls. The average anthropometric parameters of the full-term newborn in the city of Douala were: average weight: 3305 grams, average height: 49.8 centimeters, average head circumference: 34.6 centimeters, average upper arm circumference: 11.3 centimeters, circumference average thoracic: 32.8 centimeters. The percentile distribution showed a 10th percentile at 2656 grams and a 90th percentile at 3966 grams for weight defining the limits for small-for-gestational-age neonates and macrosomes. Conclusion: The anthropometric data of the full-term newborn in the city of Douala were: an average weight of 3305.4 grams, an average height of 49.8 centimeters, an average head circumference of 34.2 centimeters, an average upper arm circumference of 11.3 centimeters, and an average thoracic circumference of 32.8 centimeters with higher valuesin male newborns.

Long-Term Mortality of Children with Congenital Heart Disease Admitted to the Departmental University Hospital of Borgou/Alibori from 2011 to 2022

Background: Congenital heart disease is a public health issue due to its incidence and mortality rate. The aim of this study was to investigate the long-term mortality of children with congenital heart disease admitted to the Departmental University Hospital of Borgou/Alibori (CHUD-B/A) from 2011 to 2022. Methods: This descriptive longitudinal study with analytical aims covered 11 years (April 1, 2011 to December 31, 2022). It consisted of a review of the records of children under 15 years of age with echocardiographically confirmed congenital heart disease. This was followed by an interview with the parents to assess the children’s current condition. Data were entered using Kobocollect software and analyzed using R Studio 4.2.2. software. Results: A total of 143 complete files were retained. The median age at diagnosis was 14 months (IIQ: Q1 = 4;Q3 = 60) with a range of 2 days and 175 months, and the sex-ratio (M/F) was 0.96. Left-to-right shunts were the most frequent cardiopathy group (62.9%). Only 35 children (24.5%) benefited from restorative treatment. The mortality rate was 31.5%. Median survival under the maximum bias assumption was 114 months and 216 months under the assumption of minimum bias. Survival was significantly better in children with right-to-left shunts (p = 0.0049) under the assumption of minimum bias. The death risk factors were: age at diagnosis less than 12 months (aHR = 7.58;95% CI = 3.36 - 17.24;p Conclusion: The long-term mortality of congenital heart disease is high and favoured by the absence of restorative treatment. Local correction of congenital heart disease and medical follow-up will help to reduce this mortality.

A Trajectory Privacy Protection Method to Resist Long-Term Observation Attacks

Users face the threat of trajectory privacy leakage when using location-based service applications, especially when their behavior is collected and stored for a long period of time. This accumulated information is exploited by opponents, greatly increasing the risk of trajectory privacy leakage. This attack method is called a long-term observation attack. On the premise of ensuring lower time overhead and higher cache contribution rate, the existing methods cannot utilize cache to answer subsequent queries while also resisting long-term observation attacks. So this article proposes a trajectory privacy protection method to resist long-term observation attacks. This method combines caching technology and improves the existing differential privacy mechanism, while incorporating randomization factors that are difficult for attackers to recognize after long-term observation to enhance privacy. Search for locations in the cache of both the mobile client and edge server that can replace the user’s actual location. If there are replacement users in the cache, the query results can be obtained more quickly. Simultaneously obfuscating the spatiotemporal correlation of actual trajectories by generating confusion regions. If it does not exist, the obfuscated location generation method that resists long-term observation attacks is executed to generate the real anonymous area and send it to the service provider. The above steps can comprehensively protect the user’s trajectory privacy. The experimental results show that this method can protect user trajectories from long-term observation attacks while ensuring low time overhead and a high cache contribution rate.

Photoelectric State with Long-Term Relaxation in CdTe:(Ag, Cu, Cd) and Sb2Se3:Se Photovoltaic Films

The results of an experimental study of long-term relaxation of the photoelectret state of polycrystalline CdTe:(Ag, Cu, Cd) and Sb2Se3:Se films with an anomalous photovoltaic property are presented. In such films, the residual photovoltage is caused by the separation of photocarriers by the built-in electrostatic field of the near-surface region of space charges and their asymmetric capture by deep levels of impurities or complexes, including impurity atoms and intrinsic defects, both in the bulk and on the surface of crystal grains. It has been shown that in activated films, a two-step exponential temporary relaxation of the initial photovoltage of the order of VAPV ≈ (500-600) V is detected, and only 10% of it experiences long-term relaxation (t ≈ 100-120 min).

基于改进TF-IDF算法的毕业生就业推荐算法研究

针对传统就业推荐算法不能够对每一个毕业生进行精准的推荐的局限性,论文提出一种结合TF-IDF算法和K-means++算法的双向推荐系统,一方面对毕业生信息使用K-means++算法进行聚类,对新用户根据其初始信息与行为信息进行用户画像建模,并计算与往届毕业生的相似度;另一方面使用TF-IDF算法对各个招聘网站所发布的招聘信息中的关键词进行统计转换词频等操作。实验结果表明,该双向就业推荐系统比起之前单向就业推荐提高了毕业生就业推荐的满意度,提升推荐效率。

基于TF-IDF和TextRank结合的中文文本关键词提取方法——以体育新闻为例

利用文本挖掘技术进行体育热点分析,可以为体育领域的发展提供更多有用的信息。文中提出了一种基于TF-IDF(Term Frequency-Inverse Document Frequency,词频-逆文档频率)和TextRank(文本排序)的中文文本关键词提取方法,该方法首先采用分词、去除停用词等对文本进行预处理;其次使用TF-IDF算法计算每个词的重要性并进行归一化处理,同时使用TextRank算法权衡单词之间的关系并计算每个单词的得分以进行归一化处理;最后将TF-IDF值和TextRank得分进行加权和得到每个词的综合权重值,最终获得权重值最高的N个关键词。应用TF-IDF和TextRank结合的方法在F1值上选择5个关键词时取得了更好的结果,相较于只使用TF-IDF方法或TextRank方法,其关键词提取准确率分别提高约40%和32%。该方法有效提高了关键词提取的准确性和提取效率。

基于TF-IDF和VOSviewer的我国应急救援现状可视化分析

为有效利用消防救援队伍的实战记录资料挖掘应急救援战例成功经验,结合词频-逆文档频率(TF-IDF)算法和VOSviewer文献可视化分析技术,构建战例资料分析模型,分析战例成功与失败的共性规律和特点,总结我国应急救援现状及发展趋势。模型以2007—2019年间共185起应急救援典型战例为数据库,按照自然灾害、交通事故、建筑坍塌、危化品泄漏、火灾扑救等应急救援行动类型展开分析。结果表明:我国应急救援行动的影响因素主要表现在人(救援队伍)、机(装备技术)、环(环境)、管(管理)4个方面。其中,环境因素的影响几乎都是负面的,其他3个因素均有正负面影响。此外,不同应急救援行动类型的主导影响因素存在差异,自然灾害突出“机”;交通事故突出“管”;建筑坍塌突出“机”“环”;危化品泄漏在“人机环管”4个方面均有突出问题;火灾救援突出“机”。

融合条件熵和TF-IDF的过采样方法

针对非均衡数据带来的分类器对少数类样本学习不充分的问题,提出融合条件熵和TF-IDF的过采样方法。该方法首先指定参数,组合数据特征,然后计算每种组合方式下的条件熵,判断每种组合条件下类的不确定性,同时为了避免低词频带来的噪音数据,将条件熵结果乘上1/TF-IDF因子,再将结果按升序排序,最后结合参数选定过采样依据的特征组合,用以构造新数据,使正负样本平衡。将所提方法在7个不均衡数据集上进行实验仿真,结果表明,所提方法比其他方法在F-measure、G-mean和AUC等评价指标上均有一定提高。

一种结合TF-IDF和Simhash的科技项目文本相似性度量方法

为了提高科技项目文本相似性度量的准确性和性能,将TF-IDF和Simhash相结合,提出了一种新的科技项目文本相似性度量方法。首先,该方法对科技项目文本进行预处理得到词项集合,再使用TF-IDF计算词项集合中每个词项的权重值,并选取具有较高权重值的重要词项;其次,使用Simhash把重要词项映射为固定长度的二进制串,并求和得到文本的Simhash签名;最后,使用汉明距离计算两个Simhash签名间的相似性。实验结果表明,所提方法在查准率、召回率和F度量值方面优于传统的Simhash算法和TF-IDF方法。

Carbon sequestration rate,nitrogen use efficiency and rice yield responses to long-term substitution of chemical fertilizer by organic manure in a rice–rice cropping system

Combined application of chemical fertilizers with organic amendments was recommended as a strategy for improving yield,soil carbon storage,and nutrient use efficiency.However,how the long-term substitution of chemical fertilizer with organic manure affects rice yield,carbon sequestration rate(CSR),and nitrogen use efficiency(NUE)while ensuring environmental safety remains unclear.This study assessed the long-term effect of substituting chemical fertilizer with organic manure on rice yield,CSR,and NUE.It also determined the optimum substitution ratio in the acidic soil of southern China.The treatments were:(i)NPK0,unfertilized control;(ii)NPK1,100%chemical nitrogen,phosphorus,and potassium fertilizer;(iii)NPKM1,70%chemical NPK fertilizer and 30%organic manure;(iv)NPKM2,50%chemical NPK fertilizer and 50%organic manure;and(v)NPKM3,30%chemical NPK fertilizer and 70%organic manure.Milk vetch and pig manure were sources of manure for early and late rice seasons,respectively.The result showed that SOC content was higher in NPKM1,NPKM2,and NPKM3 treatments than in NPK0 and NPK1 treatments.The carbon sequestration rate increased by 140,160,and 280%under NPKM1,NPKM2,and NPKM3 treatments,respectively,compared to NPK1 treatment.Grain yield was 86.1,93.1,93.6,and 96.5%higher under NPK1,NPKM1,NPKM2,and NPKM3 treatments,respectively,compared to NPK0 treatment.The NUE in NPKM1,NPKM2,and NPKM3 treatments was higher as compared to NPK1 treatment for both rice seasons.Redundancy analysis revealed close positive relationships of CSR with C input,total N,soil C:N ratio,catalase,and humic acids,whereas NUE was closely related to grain yield,grain N content,and phenol oxidase.Furthermore,CSR and NUE negatively correlated with humin acid and soil C:P and N:P ratios.The technique for order of preference by similarity to ideal solution(TOPSIS)showed that NPKM3 treatment was the optimum strategy for improving CSR and NUE.Therefore,substituting 70%of chemical fertilizer with organic manure could be the best management option for increasing CSR and NUE in the paddy fields of southern China.

基于差异化建模与TF-IDF算法的城市功能区识别及混合度测算

基于POI性质、特点的不同,本文构建了一个融合统计分析法、核密度分析法的城市功能区定量识别模型,有效地识别出了广州市中心四区的功能区类型。利用耦合TF-IDF算法和信息熵算法测算城市功能混合度并进行面积加权,使得城市功能混合度测算更加符合实际情况。研究结果表明:广州市中心城区呈现出混合用地为主的特征,总体混合程度高。混合用地主要分布在研究区中心,单一类型用地零星分布在研究区外围,呈现出显著的“核心-外围”式的圈层化分布格局。城市功能混合度呈现“中心高,四周低”“多中心,组团式”、空间梯度差异显著的分布特征,并且功能混合程度与发展水平有一定正相关关系;功能分区结果与混合密度情况分布较为一致,表明本研究方法可行、研究结果合理。