Set Of Annotations Research Articles

ClusterMine: A knowledge-integrated clustering approach based on expression profiles of gene sets.

Clustering analysis of gene expression data is essential for understanding complex biological data, and is widely used in important biological applications such as the identification of cell subpopulations and disease subtypes. In commonly used methods such as hierarchical clustering (HC) and consensus clustering (CC), holistic expression profiles of all genes are often used to assess the similarity between samples for clustering. While these methods have been proven successful in identifying sample clusters in many areas, they do not provide information about which gene sets (functions) contribute most to the clustering, thus limiting the interpretability of the resulting cluster. We hypothesize that integrating prior knowledge of annotated gene sets would not only achieve satisfactory clustering performance but also, more importantly, enable potential biological interpretation of clusters. Here we report ClusterMine, an approach that identifies clusters by assessing functional similarity between samples through integrating known annotated gene sets in functional annotation databases such as Gene Ontology. In addition to the cluster membership of each sample as provided by conventional approaches, it also outputs gene sets that most likely contribute to the clustering, thus facilitating biological interpretation. We compare ClusterMine with conventional approaches on nine real-world experimental datasets that represent different application scenarios in biology. We find that ClusterMine achieves better performances and that the gene sets prioritized by our method are biologically meaningful. ClusterMine is implemented as an R package and is freely available at: www.genemine.org/clustermine.php.

Explaining the semantics capturing capability of scene graph generation models

Deep neural network is a effective way for scene graph generation tasks. However, it also makes the scene graph generation models difficult to explain. For instance, the current standard metric cannot explain how capable neural network models are of capturing the semantics of relations. In this paper, we try to understand the semantics capturing capability of scene graph generation models based on three types of metrics: conformance recall, violation recall, and non-violation recall, which measure semantic properties of relations that are reflected by triples in scene graph generated by models. Evaluation of these metrics on three representative state-of-the-art scene graph generation models based on deep neural network in Visual Genome dataset shows that the proposed metrics can effectively explain the capability of models to capture different semantic properties and identify design problems in models. By extending the Visual Genome dataset with different sets of additional annotations, these metrics can also explaining whether the semantics capturing capability of deep neural network models can be improved by data enhancement.

Extracting chemical reactions from text using Snorkel

BackgroundEnzymatic and chemical reactions are key for understanding biological processes in cells. Curated databases of chemical reactions exist but these databases struggle to keep up with the exponential growth of the biomedical literature. Conventional text mining pipelines provide tools to automatically extract entities and relationships from the scientific literature, and partially replace expert curation, but such machine learning frameworks often require a large amount of labeled training data and thus lack scalability for both larger document corpora and new relationship types.ResultsWe developed an application of Snorkel, a weakly supervised learning framework, for extracting chemical reaction relationships from biomedical literature abstracts. For this work, we defined a chemical reaction relationship as the transformation of chemical A to chemical B. We built and evaluated our system on small annotated sets of chemical reaction relationships from two corpora: curated bacteria-related abstracts from the MetaCyc database (MetaCyc_Corpus) and a more general set of abstracts annotated with MeSH (Medical Subject Headings) term Bacteria (Bacteria_Corpus; a superset of MetaCyc_Corpus). For the MetaCyc_Corpus, we obtained 84% precision and 41% recall (55% F1 score). Extending to the more general Bacteria_Corpus decreased precision to 62% with only a four-point drop in recall to 37% (46% F1 score). Overall, the Bacteria_Corpus contained two orders of magnitude more candidate chemical reaction relationships (nine million candidates vs 68,0000 candidates) and had a larger class imbalance (2.5% positives vs 5% positives) as compared to the MetaCyc_Corpus. In total, we extracted 6871 chemical reaction relationships from nine million candidates in the Bacteria_Corpus.ConclusionsWith this work, we built a database of chemical reaction relationships from almost 900,000 scientific abstracts without a large training set of labeled annotations. Further, we showed the generalizability of our initial application built on MetaCyc documents enriched with chemical reactions to a general set of articles related to bacteria.

Identification of potential molecular mechanisms of radiation pneumonitis development in non-small-cell lung cancer treatment by data mining

Introduction: Radiation pneumonitis (RP) is the most significant dose-limiting toxicity in patients receiving thoracic radiotherapy. The underlying mechanisms of RP are still inconclusive. Our objective was to determine the genes and molecular pathways associated with RP using computational tools and publicly available data. Methods: RP-associated genes were determined by text mining, and the intersection of the two gene sets was selected for Gene Ontology analysis using the GeneCodis program. Protein-protein interaction network analysis was performed using STRINGdb to identify the final genes. Results: Our analysis identified 256 genes related to RP with text mining. The enriched biological process annotations resulted in 47 sets of annotations containing a total of 156 unique genes. KEGG analysis of the enriched pathways identified 24 pathways containing a total of 41 unique genes. The protein-protein interaction analysis yielded 23 genes (mostly the PI3K family). Conclusion: Gene discovery using in silico text mining and pathway analysis tools can facilitate the identification of the underlying mechanisms of RP.

What am I testing and where? Comparing testing procedures based on lightweight requirements annotations

ContextThe testing of software-intensive systems is performed in different test stages each having a large number of test cases. These test cases are commonly derived from requirements. Each test stages exhibits specific demands and constraints with respect to their degree of detail and what can be tested. Therefore, specific test suites are defined for each test stage. In this paper, the focus is on the domain of embedded systems, where, among others, typical test stages are Software- and Hardware-in-the-loop.ObjectiveMonitoring and controlling which requirements are verified in which detail and in which test stage is a challenge for engineers. However, this information is necessary to assure a certain test coverage, to minimize redundant testing procedures, and to avoid inconsistencies between test stages. In addition, engineers are reluctant to state their requirements in terms of structured languages or models that would facilitate the relation of requirements to test executions.MethodWith our approach, we close the gap between requirements specifications and test executions. Previously, we have proposed a lightweight markup language for requirements which provides a set of annotations that can be applied to natural language requirements. The annotations are mapped to events and signals in test executions. As a result, meaningful insights from a set of test executions can be directly related to artifacts in the requirements specification. In this paper, we use the markup language to compare different test stages with one another.ResultsWe annotate 443 natural language requirements of a driver assistance system with the means of our lightweight markup language. The annotations are then linked to 1300 test executions from a simulation environment and 53 test executions from test drives with human drivers. Based on the annotations, we are able to analyze how similar the test stages are and how well test stages and test cases are aligned with the requirements. Further, we highlight the general applicability of our approach through this extensive experimental evaluation.ConclusionWith our approach, the results of several test levels are linked to the requirements and enable the evaluation of complex test executions. By this means, practitioners can easily evaluate how well a systems performs with regards to its specification and, additionally, can reason about the expressiveness of the applied test stage.

The gene-rich genome of the scallop Pecten maximus.

BackgroundThe king scallop, Pecten maximus, is distributed in shallow waters along the Atlantic coast of Europe. It forms the basis of a valuable commercial fishery and plays a key role in coastal ecosystems and food webs. Like other filter feeding bivalves it can accumulate potent phytotoxins, to which it has evolved some immunity. The molecular origins of this immunity are of interest to evolutionary biologists, pharmaceutical companies, and fisheries management.FindingsHere we report the genome assembly of this species, conducted as part of the Wellcome Sanger 25 Genomes Project. This genome was assembled from PacBio reads and scaffolded with 10X Chromium and Hi-C data. Its 3,983 scaffolds have an N50 of 44.8 Mb (longest scaffold 60.1 Mb), with 92% of the assembly sequence contained in 19 scaffolds, corresponding to the 19 chromosomes found in this species. The total assembly spans 918.3 Mb and is the best-scaffolded marine bivalve genome published to date, exhibiting 95.5% recovery of the metazoan BUSCO set. Gene annotation resulted in 67,741 gene models. Analysis of gene content revealed large numbers of gene duplicates, as previously seen in bivalves, with little gene loss, in comparison with the sequenced genomes of other marine bivalve species.ConclusionsThe genome assembly of P. maximus and its annotated gene set provide a high-quality platform for studies on such disparate topics as shell biomineralization, pigmentation, vision, and resistance to algal toxins. As a result of our findings we highlight the sodium channel gene Nav1, known to confer resistance to saxitoxin and tetrodotoxin, as a candidate for further studies investigating immunity to domoic acid.

Non-homology-based prediction of gene functions in maize (Zea mays ssp. mays).

Advances in genome sequencing and annotation have eased the difficulty of identifying new gene sequences. Predicting the functions of these newly identified genes remains challenging. Genes descended from a common ancestral sequence are likely to have common functions. As a result, homology is widely used for gene function prediction. This means functional annotation errors also propagate from one species to another. Several approaches based on machine learning classification algorithms were evaluated for their ability to accurately predict gene function from non-homology gene features. Among the eight supervised classification algorithms evaluated, random-forest-based prediction consistently provided the most accurate gene function prediction. Non-homology-based functional annotation provides complementary strengths to homology-based annotation, with higher average performance in Biological Process GO terms, the domain where homology-based functional annotation performs the worst, and weaker performance in Molecular Function GO terms, the domain where the accuracy of homology-based functional annotation is highest. GO prediction models trained with homology-based annotations were able to successfully predict annotations from a manually curated "gold standard" GO annotation set. Non-homology-based functional annotation based on machine learning may ultimately prove useful both as a method to assign predicted functions to orphan genes which lack functionally characterized homologs, and to identify and correct functional annotation errors which were propagated through homology-based functional annotations.

Deep Learning Method for Mandibular Canal Segmentation in Dental Cone Beam Computed Tomography Volumes

Accurate localisation of mandibular canals in lower jaws is important in dental implantology, in which the implant position and dimensions are currently determined manually from 3D CT images by medical experts to avoid damaging the mandibular nerve inside the canal. Here we present a deep learning system for automatic localisation of the mandibular canals by applying a fully convolutional neural network segmentation on clinically diverse dataset of 637 cone beam CT volumes, with mandibular canals being coarsely annotated by radiologists, and using a dataset of 15 volumes with accurate voxel-level mandibular canal annotations for model evaluation. We show that our deep learning model, trained on the coarsely annotated volumes, localises mandibular canals of the voxel-level annotated set, highly accurately with the mean curve distance and average symmetric surface distance being 0.56 mm and 0.45 mm, respectively. These unparalleled accurate results highlight that deep learning integrated into dental implantology workflow could significantly reduce manual labour in mandibular canal annotations.

Fully Automated Bone Age Assessment on Large-Scale Hand X-Ray Dataset.

Bone age assessment (BAA) is an essential topic in the clinical practice of evaluating the biological maturity of children. Because the manual method is time-consuming and prone to observer variability, it is attractive to develop computer-aided and automated methods for BAA. In this paper, we present a fully automatic BAA method. To eliminate noise in a raw X-ray image, we start with using U-Net to precisely segment hand mask image from a raw X-ray image. Even though U-Net can perform the segmentation with high precision, it needs a bigger annotated dataset. To alleviate the annotation burden, we propose to use deep active learning (AL) to select unlabeled data samples with sufficient information intentionally. These samples are given to Oracle for annotation. After that, they are then used for subsequential training. In the beginning, only 300 data are manually annotated and then the improved U-Net within the AL framework can robustly segment all the 12611 images in RSNA dataset. The AL segmentation model achieved a Dice score at 0.95 in the annotated testing set. To optimize the learning process, we employ six off-the-shell deep Convolutional Neural Networks (CNNs) with pretrained weights on ImageNet. We use them to extract features of preprocessed hand images with a transfer learning technique. In the end, a variety of ensemble regression algorithms are applied to perform BAA. Besides, we choose a specific CNN to extract features and explain why we select that CNN. Experimental results show that the proposed approach achieved discrepancy between manual and predicted bone age of about 6.96 and 7.35 months for male and female cohorts, respectively, on the RSNA dataset. These accuracies are comparable to state-of-the-art performance.

What and where: A context-based recommendation system for object insertion

We propose a novel problem revolving around two tasks: (i) given a scene, recommend objects to insert, and (ii) given an object category, retrieve suitable background scenes. A bounding box for the inserted object is predicted in both tasks, which helps downstream applications such as semiautomated advertising and video composition. The major challenge lies in the fact that the target object is neither present nor localized in the input, and furthermore, available datasets only provide scenes with existing objects. To tackle this problem, we build an unsupervised algorithm based on object-level contexts, which explicitly models the joint probability distribution of object categories and bounding boxes using a Gaussian mixture model. Experiments on our own annotated test set demonstrate that our system outperforms existing baselines on all sub-tasks, and does so using a unified framework. Future extensions and applications are suggested.

EDGE20: A Cross Spectral Evaluation Dataset for Multiple Surveillance Problems.

Surveillance-related datasets that have been released in recent years focus only on one specific problem at a time (e.g., pedestrian detection, face detection, or face recognition), while most of them were collected using visible spectrum (VIS) cameras. Even though some cross-spectral datasets were presented in the past, they were acquired in a constrained setup, which limited the performance of methods for the aforementioned problems under a cross-spectral setting. This work introduces a new dataset, named EDGE19, that can be used in addressing the problems of pedestrian detection, face detection, and face recognition in images captured using trail cameras under the VIS and NIR spectra. Data acquisition was performed in an outdoor environment, during both day and night, under unconstrained acquisition conditions. The collection of images is accompanied by a rich set of annotations, consisting of person and facial bounding boxes, unique subject identifiers, and labels that characterize facial images as frontal, profile, or back faces. Moreover, the performance of several state-of-the-art methods was evaluated for each of the scenarios covered by our dataset. The baseline results we obtained highlight the difficulty of current methods in the tasks of cross-spectral pedestrian detection, face detection, and face recognition due to unconstrained conditions, including low resolution, pose variation, illumination variation, occlusions, and motion blur.

The Woman at the Well and the Limits of Biblical Scholarship: Maximus the Confessor’s Scholion on Thal. 41

Abstract In 2018 Maximus the Confessor’s premier work on biblical hermeneutics, the Responses to Thalassius, finally appeared in English translation. Following its original publication in the early 630 s, Maximus reissued the Responses in a second edition, to which he appended a dedicated prologue and his so-called scholia, an extensive set of annotations or footnoted clarifications. In both Maximus’s prologue and in the reception history of the Responses, these scholia were regarded as intrinsic to the integrity of the whole work. This article focuses on scholion 1 to Thal. 41, in which Maximus comments on the number of husbands belonging to the Woman at the Well in John 4, and why Jesus’ conversation with her took place when and where it did. It treats the scholion as a test case to see whether, how, and to what extent it further enlightens the reader as to the meaning of Maximus’s initial commentary, as he says it should. It argues that the scholion crucially qualifies several insights raised by Maximus in his original response, touching on his anagogical reading of Scripture, the progressive character of human history towards a culminating salvific goal, the limits of learning and discursive reason, and the role of faith and grace in receiving deifying wisdom.

The MatchNMingle Dataset: A Novel Multi-Sensor Resource for the Analysis of Social Interactions and Group Dynamics In-the-Wild During Free-Standing Conversations and Speed Dates

We present MatchNMingle , a novel multimodal/multisensor dataset for the analysis of free-standing conversational groups and speed-dates in-the-wild. MatchNMingle leverages the use of wearable devices and overhead cameras to record social interactions of 92 people during real-life speed-dates, followed by a cocktail party. To our knowledge, MatchNMingle has the largest number of participants, longest recording time and largest set of manual annotations for social actions available in this context in a real-life scenario. It consists of 2 hours of data from wearable acceleration, binary proximity, video, audio, personality surveys, frontal pictures and speed-date responses. Participants’ positions and group formations were manually annotated; as were social actions (eg. speaking, hand gesture) for 30 minutes at 20 FPS making it the first dataset to incorporate the annotation of such cues in this context. We present an empirical analysis of the performance of crowdsourcing workers against trained annotators in simple and complex annotation tasks, founding that although efficient for simple tasks, using crowdsourcing workers for more complex tasks like social action annotation led to additional overhead and poor inter-annotator agreement compared to trained annotators (differences up to 0.4 in Fleiss’ Kappa coefficients). We also provide example experiments of how MatchNMingle can be used.

Bach and Mozart: Essays on the Enigma of Genius by Robert L. Marshall

Reviewed by: Bach and Mozart: Essays on the Enigma of Genius by Robert L. Marshall Marshall Brown (bio) Robert L. Marshall. Bach and Mozart: Essays on the Enigma of Genius (Rochester, NY: University of Rochester Press, 2019). xxi, 330 pp. Robert L. Marshall’s Bach and Mozart: Essays on the Enigma of Genius collects essays originally published since 1990, as a sequel to the 1989 collection The Music of Johann Sebastian Bach, which assembled essays from 1973 forward.1 In both books, some of the essays have brief added postscripts to bring them up to date, and the new essays are also lightly adapted to avoid anachronism and to incorporate cross-references. Marshall is surely well known to readers of this journal, and at least two of the essays from the earlier book, “Bach the Progressive” and “On Bach’s Universality,” are justifiably classics. There are no major departures in the new book, but there are some significant shifts in approach. Most of the essays in Marshall’s earlier book are philological, concerned (as the section titles have it) with evidence for Bach’s compositional process, with “Questions of Authenticity and Chronology,” and with performance practice. The earlier essays in the new book continue in that vein. Brief essays on the Wilhelm Friedrich Bach and Anna Magdalena Bach notebooks and on Mozart’s unfinished compositions review the evidence meticulously and draw modest conclusions, working toward serious reflection (e.g., “it appears that the creative act was always driven forward, step by step, along the unfolding form,” 251). While the notebook essay ends in hyperbole (“The historical value of the notebooks .. . can hardly be exaggerated,” 37), I welcome the impulse to employ the essayist’s encyclopedic knowledge toward some general reflections. Conversely, the longest essay, a survey of Bach’s keyboard music, is in the vein of the earlier volume’s survey of his music for flute, a set of learned and judicious annotations with only a brief final gesture toward a larger argument. Judicious learning has always been Marshall’s forte. Some triumphs here are: 1) a brief compromise with “my good friend [Joshua] [End Page 306] Rifkin” (115) on the size of Bach’s chorus (Marshall concedes that probably the choral movements were sometimes performed with Rifkin’s single voices, but surely also sometimes with larger groups); 2) a defense of the St. John Passion that leaves room for Jewish discomfort with the antisemitic Gospel text (Marshall himself is Jewish); 3) a welcoming review of Peter Shaffer’s Amadeus on the occasion of the release of the film version (yes, the facts are distorted to accentuate Salieri’s presumed perspective, but it’s good entertainment and aroused substantial interest in Mozart); and especially 4) the book’s liveliest read, a wry survey of the scant information leading to critics’ innuendos about Johann Christian Bach’s sexuality. Earlier, Marshall dug into archives and scrutinized manuscripts. Bach and Mozart chiefly draws on book stacks, with an encyclopedic command of information old and new, unruffled by rare errors. Adorno did not coin the term Spätstil (149), but that hardly matters. Somewhat more consequential is the claim of affinity that “Bach, like Luther, was a native Thuringian” (54). Luther’s birthplace, Eisleben, in the county of Mansfeld, is in the modern state of Sachsen-Anhalt, near the border of Thuringia but not in it, now or then, and while Luther was exiled to Eisenach in Thuringia for one crucial year, he lived chiefly in Wittenberg, in the heart of Saxony. And I don’t know of any warrant for the proposal in the Luther chapter that “one would like to think that for the formally gifted Johann Sebastian Bach [aufführen didn’t just mean perform but] also implied ‘to create’” (38). This last example is an instance of the book’s core mode, a somewhat loose weave of persuasive conjecture supported by endless information, not damaged by the very occasional out-of-place stitch. In a given paragraph—and there are many such—one may find: “if we are willing to lend any credence at all .. . and if we are willing to indulge a degree of plausible speculation in the absence...

Автоматическое тегирование изображений одежды

This paper presents a computer vision clothing auto-tagging algorithm. Tagging is highly demanded in e-commerce as a tool to create a rich uniform set of annotations. The annotations improve catalog organization, statistics, and can be used for interactive catalog search by consumer photos. The proposed algorithm predicts length, design, and color attributes for an arbitrary number of clothing items in an image. The modular structure of the proposed system allows reconfiguration for other sets of tags and tagging tasks not related to clothing.

JHU-CROWD++: Large-Scale Crowd Counting Dataset and A Benchmark Method.

We introduce a new large scale unconstrained crowd counting dataset (JHU-CROWD++) that contains "4,372" images with "1.51 million" annotations. In comparison to existing datasets, the proposed dataset is collected under a variety of diverse scenarios and environmental conditions. Specifically, the dataset includes several images with weather-based degradations and illumination variations, making it a very challenging dataset. Additionally, the dataset consists of a rich set of annotations at both image-level and head-level. Several recent methods are evaluated and compared on this dataset. The dataset can be downloaded from http://www.crowd-counting.com. Furthermore, we propose a novel crowd counting network that progressively generates crowd density maps via residual error estimation. The proposed method uses VGG16 as the backbone network and employs density map generated by the final layer as a coarse prediction to refine and generate finer density maps in a progressive fashion using residual learning. Additionally, the residual learning is guided by an uncertainty-based confidence weighting mechanism that permits the flow of only high-confidence residuals in the refinement path. The proposed Confidence Guided Deep Residual Counting Network (CG-DRCN) is evaluated on recent complex datasets, and it achieves significant improvements In errors.

Towards Describing Visual Explanation using Machine Learning

Current available visible explanation generating systems research to easily absolve a class prediction. Still, they may additionally point out visible parameters attribute which replicate a strong category prior, though the proof may additionally not clearly be in the pic. This is specifically regarding as alternatively such marketers fail in constructing have confidence with human users. We proposed our own version, which makes a speciality of the special places of house of the seen item, together predicts the category label & interprets why the expected label is proper for the image. The machine proposes to annotate the images automatically using the Markov cache model. To annotate images, principles are represented as states through the usage of Hidden Markov model. The model parameters were estimated as part of a set of images and manual annotations. This is a great collection of checks, albeit automatically, with the possibility a posteriori of the concepts presented in her.

YAP1 Mediates Resistance to MEK1/2 Inhibition in Neuroblastomas with Hyperactivated RAS Signaling.

Relapsed neuroblastomas are enriched with activating mutations of the RAS-MAPK signaling pathway. The MEK1/2 inhibitor trametinib delays tumor growth but does not sustain regression in neuroblastoma preclinical models. Recent studies have implicated the Hippo pathway transcriptional coactivator protein YAP1 as an additional driver of relapsed neuroblastomas, as well as a mediator of trametinib resistance in other cancers. Here, we used a highly annotated set of high-risk neuroblastoma cellular models to modulate YAP1 expression and RAS pathway activation to test whether increased YAP1 transcriptional activity is a mechanism of MEK1/2 inhibition resistance in RAS-driven neuroblastomas. In NLF (biallelic NF1 inactivation) and SK-N-AS (NRAS Q61K) cell lines, trametinib caused a near-complete translocation of YAP1 protein into the nucleus. YAP1 depletion sensitized neuroblastoma cells to trametinib, while overexpression of constitutively active YAP1 protein induced trametinib resistance. Mechanistically, significant enhancement of G1-S cell-cycle arrest, mediated by depletion of MYC/MYCN and E2F transcriptional output, sensitized RAS-driven neuroblastomas to trametinib following YAP1 deletion. These findings underscore the importance of YAP activity in response to trametinib in RAS-driven neuroblastomas, as well as the potential for targeting YAP in a trametinib combination. SIGNIFICANCE: High-risk neuroblastomas with hyperactivated RAS signaling escape the selective pressure of MEK inhibition via YAP1-mediated transcriptional reprogramming and may be sensitive to combination therapies targeting both YAP1 and MEK.

Characterizing and automatically detecting smooth pursuit in a large-scale ground-truth data set of dynamic natural scenes.

Eye movements are fundamental to our visual experience of the real world, and tracking smooth pursuit eye movements play an important role because of the dynamic nature of our environment. Static images, however, do not induce this class of eye movements, and commonly used synthetic moving stimuli lack ecological validity because of their low scene complexity compared to the real world. Traditionally, ground truth data for pursuit analyses with naturalistic stimuli are obtained via laborious hand-labelling. Therefore, previous studies typically remained small in scale. We here present the first large-scale quantitative characterization of human smooth pursuit. In order to achieve this, we first provide a methodological framework for such analyses by collecting a large set of manual annotations for eye movements in dynamic scenes and by examining the bias and variance of human annotators. To enable further research on even larger future data sets, we also describe, improve, and thoroughly analyze a novel algorithm to automatically classify eye movements. Our approach incorporates unsupervised learning techniques and thus demonstrates improved performance with the addition of unlabelled data. The code and data related to our manual and automated eye movement annotation are publicly available via https://web.gin.g-node.org/ioannis.agtzidis/gazecom_annotations/.

Incorporating Human Domain Knowledge in 3-D LiDAR-Based Semantic Segmentation

This article studies semantic segmentation using 3D LiDAR data. Popular deep learning methods applied for this task require a large number of manual annotations to train the parameters. We propose a new method that makes full use of the advantages of traditional methods and deep learning methods via incorporating human domain knowledge into the neural network model to reduce the demand for large numbers of manual annotations and improve the training efficiency. We first pretrain a model with autogenerated samples from a rule-based classifier so that human knowledge can be propagated into the network. Based on the pretrained model, only a small set of annotations is required for further fine-tuning. Quantitative experiments show that the pretrained model achieves better performance than random initialization in almost all cases; furthermore, our method can achieve similar performance with fewer manual annotations.