A multitask transfer learning solution | Entity type features: We hypothesize that the entity types and subtypes of the relation arguments are also more likely to be associated with specific relation types. |
A multitask transfer learning solution | We refer to the set of features that contain the entity type or subtype of an argument as “arg-NE” features. |
A multitask transfer learning solution | 4.4 Imposing entity type constraints |
Conclusions and future work | In this paper, we applied multitask transfer learning to solve a weakly-supervised relation extraction problem, leveraging both labeled instances of auxiliary relation types and human knowledge including hypotheses on feature generality and entity type constraints. |
Conclusions and future work | We also leveraged additional human knowledge about the target relation type in the form of entity type constraints. |
Conclusions and future work | Experiment results on the ACE 2004 data show that the multitask transfer learning method achieves the best performance when we combine human guidance with automatic general feature selection, followed by imposing the entity type constraints. |
Experiments | Finally, TL-NE builds on top of TL-comb and uses the entity type constraints to refine the predictions. |
Introduction | ditional human knowledge about the entity type constraints on the relation arguments, which can usually be derived from the definition of a relation type. |
Task definition | For example, we may be given the entity type restrictions on the two relation arguments. |
Task definition | Nodes that represent the arguments are also labeled with the entity type , subtype and mention type as defined by ACE. |
Experiments | ACE-2004 defines 7 coarse-grained entity types , each of which are then refined into 43 fine- |
Experiments | grained entity types . |
Experiments | Using the ACE data annotated with mentions and predefined entity types, we build a fine-grained mention entity typing (MET) classifier to disambiguate between 44 labels (43 fine-grained and a null label to indicate not a mention). |
Introduction | We describe our mention entity typing system in Section 4 and features for the RE system in Section 5. |
Mention Extraction System | These mention candidates are then fed to our mention entity typing (MET) classifier for type prediction (more details in Section 6.3). |
Relation Extraction Framework | Abbreviations: Lm: predicted entity label for mention m using the mention entity typing (MET) classifier described in Section 4; PM ET: prediction probability according to the MET classifier; 75: used for thresholding. |
Relation Extraction Framework | In (Roth and Yih, 2007), the authors used entity types to constrain the (first dimensional) relation types allowed among them. |
Relation Extraction System | Due to space limitations, we refer the reader to our prior work (Chan and Roth, 2010) for the lexical, structural, mention-level, entity type , and dependency features. |
Syntactico-Semantic Structures | Table 2: Features used in our mention entity typing (MET) system. |
Syntactico-Semantic Structures | and whether they satisfy certain semantic entity type constraints. |
Algorithm 3.1 The Model | First, the algorithm enumerates all possible segments (i.e., subse-quences) of ac ending at the current token with various entity types . |
Algorithm 3.1 The Model | :c-axis and y-axis represent the input sentence and entity types , respectively. |
Algorithm 3.1 The Model | The rectangles denote segments with entity types , among which the shaded ones are three competing hypotheses ending at “1,400”. |
Background | ACE defined 7 main entity types including Person (PER), Organization (ORG), |
Features | The entity segments of 3) can be expressed as a list of triples (61, ..., em), where each segment 6, = (ui, 2),, 75,-) is a triple of start index ui, end index 21,-, and entity type 25,-. |
Features | Gazetteer features Entity type of each segment based on matching a number of gazetteers including persons, countries, cities and organizations. |
Features | Coreference consistency Coreferential entity mentions should be assigned the same entity type . |
Introduction | This problem has been artificially broken down into several components such as entity mention boundary identification, entity type classification and relation extraction. |
Abstract | We predict entity type distributions in Web search queries via probabilistic inference in graphical models that capture how entity-bearing queries are generated. |
Abstract | We jointly model the interplay between latent user intents that govern queries and unobserved entity types , leveraging observed signals from query formulations and document clicks. |
Abstract | We apply the models to resolve entity types in new queries and to assign prior type distributions over an existing knowledge base. |
Introduction | There are, however, inherent problems in the entity repositories: (a) coverage: although coverage of head entity types is often reliable, the tail can be sparse; (b) noise: created by spammers, extraction |
Introduction | We cast the task as performing probabilistic inference in a graphical model that captures how queries are generated, and then apply the model to contextually recognize entity types in new queries. |
Introduction | We show that jointly modeling user intent and entity type significantly outperforms the current state of the art on the task of entity type resolution in queries. |
Joint Model of Types and User Intents | We theorize that search queries are governed by a latent user intent, which in turn influences the entity types , the choice of query words, and the clicked hosts. |
Related Work | entity type with latent user intents, and by incorporating click signals. |
Experiments | Local+Type This system adds entity type features to the previous system. |
Experiments | This allows us to compare performance of using global features against entity type features. |
Experiments | To determine entity types , we link named entities to Wikipedia pages using the Wikifier (Rati-nov et al., 2011) package and extract categories from the Wikipedia page. |
Introduction | fine-grained entity types of two arguments, to handle polysemy. |
Introduction | However, such fine grained entity types come at a high cost. |
Introduction | It is difficult to discover a high-quality set of fine-grained entity types due to unknown criteria for developing such a set. |
Discussion | Except in Row 8 and Row 11, when two head nouns of entity pair were combined as semantic pair and when POS tag were combined with the entity type , the performances are decreased. |
Feature Construction | head noun, entity type , subtype, CLASS, LDC-TYPE, etc.) |
Feature Construction | In our experiment, the entity type , subtype and the head noun are used. |
Feature Construction | All the employed features are simply classified into five categories: Entity Type and Subtype, Head Noun, Position Feature, POS Tag and Omni-word Feature. |
Experiments | documents 1,303,000 entity pairs 2,017,000 (matched to Freebase) 129,000 (with entity types ) 913,000 frequent patterns 3,084 relations 24 |
Experiments | In Experiment 1, since we needed entity types for patterns, we restricted ourselves to entities matched with Freebase, which also provides entity types for entities. |
Experiments | Entity types are omitted in patterns. |
Related Work | (2011) applied a rule-based method to the problem by using popular entity types and keywords for each relation. |
Wrong Label Reduction | We define a pattern as the entity types of an entity pair2 as well as the sequence of words on the path of the dependency parse tree from the first entity to the second one. |
Wrong Label Reduction | 2If we use a standard named entity tagger, the entity types are Person, Location, and Organization. |
Entity Relevance | The examples we are interested in are in the medical domain and deal with three main entity types : PERSON, DRUG, and DISEASE, where PERSON is restricted to known physicians. |
Entity Relevance | While each of the entity types can be the target of a sentiment expression, the more interesting questions in this domain involve multiple entities, specifically, DRUG + DISEASE ("how effective is this drug for this disease? |
Experiments | In the Financial corpus, COMPANIES are used as target entities and in the medical corpus, DISEASES, DRUGS and PERSONS are the entity types that are used as target entities. |
Introduction | Another layer that we'd like to add concerns the interaction of different entity types during SA. |
Introduction | In a typical situation, there is only one entity type which is the target for SA. |
Introduction | In such cases, clearly distinguishing between the relevancy of target and non-target entities types is not essential. |
Relevance Algorithms | (2010), working in the 'ignore relevance' mode, which (1) finds and labels all entities of the target type(s); (2) resolves all corefer-ences for the target entity type (s); (3) finds and labels all sentiment expressions, regardless of their relevance; and (4) provides dependency parses for all sentences in the corpus. |
Experimental Design | In this task, a human annotator has to decide for each word whether or not it belongs to one of the entity types of interest. |
Experimental Design | Annotation of these entity types in newspaper articles is admittedly fairly easy. |
Experimental Design | Moreover, the limited number of entity types reduced the amount of participants’ training prior to the actual experiment, and positively affected the design and handling of the experimental apparatus (see below). |
Abstract | We investigate a largely unsupervised approach to learning interpretable, domain-specific entity types from unlabeled text. |
Abstract | It assumes that any common noun in a domain can function as potential entity type , and uses those nouns as hidden variables in a HMM. |
Abstract | The results suggest that it is possible to learn domain-specific entity types from unlabeled data. |
Conclusion | We evaluated an approach to learning domain-specific interpretable entity types from unlabeled data. |
Introduction | (2011) proposed an approach that uses co-occurrence patterns to find entity type candidates, and then learns their applicability to relation arguments by using them as latent variables in a first-order HMM. |
Introduction | 0 the learned entity types can be used to predict selectional restrictions with high accuracy |
Headline generation | (1): an MST is extracted from the entity pair 61, 62 (2); nodes are heuristically added to the MST to enforce grammaticality (3); entity types are recombined to generate the final patterns (4). |
Headline generation | COMBINEENTITYTYPES: Finally, a distinct pattern is generated from each possible combination of entity type assignments for the participating entities. |
Headline generation | While in many cases information about entity types would be sufficient to decide about the order of the entities in the generated sentences (e. g., “[person] married in [location]” for the entity set {ea 2 “Mr. |
Related work | Chambers and Jurafsky (2009) present an unsupervised method for learning narrative schemas from news, i.e., coherent sets of events that involve specific entity types (semantic roles). |
Related work | Similarly to them, we move from the assumptions that 1) utterances involving the same entity types within the same document (in our case, a collection of related documents) are likely describing aspects of the same event, and 2) meaningful representations of the underlying events can be learned by clustering these utterances in a principled way. |
Fact Candidates | Entity Types . |
Fact Candidates | We look up entity types in a knowledge |
Fact Candidates | In particular, we use the NELL entity typing API (Carlson et al., 2010). |
Experiments | Depending on the document category, we found some variations as to which hierarchy was learned in each setting, but we noticed that parameters starting with right and left gramtypes often produced quite good hierarchies: for instance right gramtype —> left gramtype —> same sentence —> right named entity type . |
Hierarchizing feature spaces | , 0 entity types . |
Introduction | The main question we raise is, given a set of indicators (such as grammatical types, distance between two mentions, or named entity types ), how to best partition the pool of mention pair examples in order to best discriminate coreferential pairs from non coreferential ones. |
System description | We used classical features that can be found in details in (Bengston and Roth, 2008) and (Rah-man and Ng, 2011): grammatical type and subtype of mentions, string match and substring, apposition and copula, distance (number of separating mentions/sentences/words), gender/number match, synonymy/hypemym and animacy (using WordNet), family name (based on lists), named entity types , syntactic features (gold parse) and anaphoricity detection. |
System description | As indicators we used: left and right grammatical types and subtypes, entity types , a boolean indicating if the mentions are in the same sentence, and a very coarse histogram of distance in terms of sentences. |
Detailed generative story | (c) If wdk is a named entity type (PERSON, PLACE, ORG, . |
Detailed generative story | One could also make more specific versions of any feature by conjoining it with the entity type t. |
Detailed generative story | More generally, the probability (2) may also be conditioned on other variables such as on the languages pi and sci—this leaves room for a transliteration model when 53.6 75 p.6—and on the entity type cut. |
Generative Model of Coreference | However, any topic may generate an entity type , 6. g. PERSON, which is then replaced by a specific name: when PERSON is generated, the model chooses a previous mention of any person and copies it, perhaps mutating its name.1 Alternatively, the model may manufacture |
Experimental Evaluation | It consists of 57 entity types and 79 relations. |
Experimental Evaluation | The entity types include Agent, PhysicalThing, Event, TimeLocation, Gender, and Group, each with several subtypes. |
Experimental Evaluation | LIME learned several rules that had only entity types in their bodies. |
Related Work | Unlike our system and others (Carlson et al., 2010; Doppa et al., 2010; Sorower et al., 2011) that use a predefined ontology, they automatically identify a set of entity types and relations using “open IE.” They use HOLMES (Schoenmackers et al., 2008), an inference engine based on MLNs (Domingos and Lowd, 2009) (an SRL approach that combines first-order logic and Markov networks) to infer additional facts. |
Experiments | Table 3 reports the NER performance of our method for each entity type, from which we see that our system consistently yields better F1 on all entity types than S B R. We also see that our system boosts the F1 for ORGANIZATION most significantly, reflecting the fact that a large number of organizations that are incorrectly labeled as PERSON by S B R, are now correctly recognized by our method. |
Experiments | Table 3: F1 (%) of NER on different entity types . |
Our Method | To resolve NER, we assign a label to each word in a tweet, indicating both the boundary and entity type . |
Our Method | tiflfil, respectively, refer to the same entity if and only if: 1) The two mentions share the same entity type ; 2) is a substring of film” or vise versa; ands) 23%.. = 1, z' = .27. andj = .71.--- .32, if 277%” exists. |
Experiments | YAGO is different from ACE 2004 in two aspects: there is less overlapping of topics, entity types and relation types between domains; and it has more relation mentions with 11 mentions per pair of entities on the average. |
Problem Statement | Entity Features Entity types and entity mention types are very useful for relation extraction. |
Problem Statement | use a subgraph in the relation instance graph (J iang and Zhai, 2007b) that contains only the node presenting the head word of the entity A, labeled with the entity type or entity mention types, to describe a single entity attribute. |
Problem Statement | The nodes that represent the argument are also labeled with the entity type , subtype and mention type. |
Models | Each mention 2' has been augmented with a single property node pi E {1, ..., The unary B factors encode prior knowledge about the setting of each pi; these factors may be hard (I will not refer to a plural entity), soft (such as a distribution over named entity types output by an NER tagger), or practically uniform (e. g. the last name Smith does not specify a particular gender). |
Models | Suppose that we are using named entity type as an entity-level property. |
Related Work | Their system could be extended to handle property information like we do, but our system has many other advantages, such as freedom from a pre-specified list of entity types , the ability to use multiple input clusterings, and discriminative projection of clusters. |
Contextual Preferences Models | We identify entity types using the default Lingpipe2 Named-Entity Recognizer (NER), which recognizes the types Location, Person and Organization. |
Contextual Preferences Models | A variable j has a single preferred entity type in cpv;n(t)[j], the type of its instantiation in 75. |
Experimental Settings | The Contextual Preferences for h were constructed manually: the named-entity types for cpvm(h) were set by adapting the entity types given in the guidelines to the types supported by the Ling-pipe NER (described in Section 3.2). |
Experiments | For the overall performance, we use the average Precision, Recall and F1, where the weight of each name entity type is proportional to the number of entities of that type. |
Experiments | Tables 2-5 report the results on each entity type, indicating that our method consistently yields better results on all entity types . |
Our Method | Following the common practice , we adopt a sequential labeling approach to jointly resolve these subtasks, i.e., for each word in the input tweet, a label is assigned to it, indicating both the boundary and entity type . |
Model | c has entity type 6 (person, organization, etc.) |
Model | Entity Type (x, e) |
Model | The predicates we use involve matching strings (checking if two mentions share a head word or if they are exactly the same string), matching argreement features (if the gender, number or person of pairs of NPs are the same; especially important for pronouns), the distance between mentions, and if mentions have the same entity type (i.e. |
Introduction | Features about the entity information of arguments, including: a) #TP]-#TP2: the concat of the major entity types of arguments; b) #STI-#ST2: the concat of the sub entity types of arguments; c) #MT] -#MT2: the concat of the mention types of arguments. |
Introduction | We capture the property of a node’s content using the following features: a) MB_#Num: The number of mentions contained in the phrase; b) MB_C_#Type: A feature indicates that the phrase contains a mention with major entity type #Type; c) M W_#Num: The number of words within the phrase. |
Introduction | a) #RP_Arg]Head_#Arg] Type: a feature indicates the relative position of a phrase node with argument 1’s head phrase, where #RP is the relative position (one of match, cover, within, overlap, other), and #Arg] Type is the major entity type of argument 1. |